Ch10-deeplearning-lab.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "5afc399d",
   "metadata": {},
   "source": [
    "# Deep Learning\n",
    "\n",
    "<a target=\"_blank\" href=\"https://colab.research.google.com/github/intro-stat-learning/ISLP_labs/blob/v2.2/Ch10-deeplearning-lab.ipynb\">\n",
    "<img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/>\n",
    "</a>\n",
    "\n",
    "[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/intro-stat-learning/ISLP_labs/v2.2?labpath=Ch10-deeplearning-lab.ipynb)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2f2304ae",
   "metadata": {},
   "source": [
    "In this section we  demonstrate how to fit the examples discussed\n",
    "in the text. We use the `Python`{} `torch` package, along with the\n",
    "`pytorch_lightning` package which provides utilities to simplify\n",
    "fitting and evaluating models.  This code can be impressively fast\n",
    "with certain  special processors,  such as Apple’s new M1 chip. The package is well-structured, flexible, and will feel comfortable\n",
    "to `Python`{} users. A good companion is the site\n",
    "[pytorch.org/tutorials](https://pytorch.org/tutorials/beginner/basics/intro.html).\n",
    "Much of our code is adapted from there, as well as the `pytorch_lightning` documentation. {The precise URLs at the time of writing are <https://pytorch.org/tutorials/beginner/basics/intro.html> and <https://pytorch-lightning.readthedocs.io/en/latest/>.}\n",
    "\n",
    "We start with several standard imports that we have  seen before."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "71bd6b71",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [],
   "source": [
    "import numpy as np, pandas as pd\n",
    "from matplotlib.pyplot import subplots\n",
    "from sklearn.linear_model import \\\n",
    "     (LinearRegression,\n",
    "      LogisticRegression,\n",
    "      Lasso)\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.model_selection import KFold\n",
    "from sklearn.pipeline import Pipeline\n",
    "from ISLP import load_data\n",
    "from ISLP.models import ModelSpec as MS\n",
    "from sklearn.model_selection import \\\n",
    "     (train_test_split,\n",
    "      GridSearchCV)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9b799d88",
   "metadata": {},
   "source": [
    "### Torch-Specific Imports\n",
    "There are a number of imports for `torch`. (These are not\n",
    "included with `ISLP`, so must be installed separately.)\n",
    "First we import the main library\n",
    "and essential tools used to specify sequentially-structured networks."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "3298bdf1",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "from torch import nn\n",
    "from torch.optim import RMSprop\n",
    "from torch.utils.data import TensorDataset\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "51c32fa5",
   "metadata": {},
   "source": [
    "There are several other helper packages for `torch`. For instance,\n",
    "the `torchmetrics` package has utilities to compute\n",
    "various metrics to evaluate performance when fitting\n",
    "a model. The `torchinfo` package provides a useful\n",
    "summary of the layers of a model. We use the `read_image()`\n",
    "function when loading test images in Section~\\ref{Ch13-deeplearning-lab:using-pretrained-cnn-models}.\n",
    "\n",
    "If you have not already installed the packages `torchvision`\n",
    "and `torchinfo` you can install them by running\n",
    "`pip install torchinfo torchvision`.\n",
    "We can now import from `torchinfo`.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "aeae9bd9",
   "metadata": {},
   "outputs": [],
   "source": [
    "from torchmetrics import (MeanAbsoluteError,\n",
    "                          R2Score)\n",
    "from torchinfo import summary\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1049a398",
   "metadata": {},
   "source": [
    "The package `pytorch_lightning` is a somewhat higher-level\n",
    "interface to `torch` that simplifies the specification and\n",
    "fitting of\n",
    "models by reducing the amount of boilerplate code needed\n",
    "(compared to using `torch` alone)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "a5a39702",
   "metadata": {},
   "outputs": [],
   "source": [
    "from pytorch_lightning import Trainer\n",
    "from pytorch_lightning.loggers import CSVLogger\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e1b27f4d",
   "metadata": {},
   "source": [
    "In order to reproduce results we use `seed_everything()`. We will also instruct `torch` to use deterministic algorithms\n",
    "where possible."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "11bf06a4",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Seed set to 0\n"
     ]
    }
   ],
   "source": [
    "from pytorch_lightning import seed_everything\n",
    "seed_everything(0, workers=True)\n",
    "torch.use_deterministic_algorithms(True, warn_only=True)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4bd719f8",
   "metadata": {},
   "source": [
    "We will use several datasets shipped with `torchvision` for our\n",
    "examples: a pretrained network for image classification,\n",
    "as well as some transforms used for preprocessing."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "062cd869",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "from torchvision.io import read_image\n",
    "from torchvision.datasets import MNIST, CIFAR100\n",
    "from torchvision.models import (resnet50,\n",
    "                                ResNet50_Weights)\n",
    "from torchvision.transforms import (Resize,\n",
    "                                    Normalize,\n",
    "                                    CenterCrop,\n",
    "                                    ToTensor)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7c2a507a",
   "metadata": {},
   "source": [
    "We have provided a few utilities in `ISLP` specifically for this lab.\n",
    "The `SimpleDataModule` and `SimpleModule` are simple\n",
    "versions of objects used in `pytorch_lightning`, the\n",
    "high-level module for fitting `torch` models. Although more advanced\n",
    "uses such as computing on graphical processing units (GPUs) and parallel data processing\n",
    "are possible in this module, we will not be focusing much on these\n",
    "in this lab. The `ErrorTracker` handles\n",
    "collections of targets and predictions over each mini-batch\n",
    "in the validation or test stage, allowing computation\n",
    "of the metric over the entire validation or test data set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "4a05884a",
   "metadata": {},
   "outputs": [],
   "source": [
    "from ISLP.torch import (SimpleDataModule,\n",
    "                        SimpleModule,\n",
    "                        ErrorTracker,\n",
    "                        rec_num_workers)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a6f41487",
   "metadata": {},
   "source": [
    "In addition we have included some helper\n",
    "functions to load the\n",
    "`IMDb` database, as well as a lookup that maps integers\n",
    "to particular keys in the database. We’ve included\n",
    "a slightly modified copy of the preprocessed\n",
    "`IMDb` data from `keras`, a separate package\n",
    "for fitting deep learning models. This saves us significant\n",
    "preprocessing and allows us to focus on specifying and fitting\n",
    "the models themselves."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "1474e88b",
   "metadata": {},
   "outputs": [],
   "source": [
    "from ISLP.torch.imdb import (load_lookup,\n",
    "                             load_tensor,\n",
    "                             load_sparse,\n",
    "                             load_sequential)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "86123617",
   "metadata": {},
   "source": [
    "Finally, we introduce some utility imports  not directly related to\n",
    "`torch`.\n",
    "The `glob()` function from the `glob` module is used\n",
    "to find all files matching wildcard characters, which we will use\n",
    "in our example applying the `ResNet50` model\n",
    "to some of our own images.\n",
    "The `json` module will be used to load\n",
    "a JSON file for looking up classes to identify the labels of the\n",
    "pictures in the `ResNet50` example."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "0100ff28",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [],
   "source": [
    "from glob import glob\n",
    "import json\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b0f81cea",
   "metadata": {},
   "source": [
    "## Single Layer Network on Hitters Data\n",
    "We start by fitting the models in Section~\\ref{Ch13:sec:when-use-deep} on the `Hitters` data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "ef59d53a",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "Hitters = load_data('Hitters').dropna()\n",
    "n = Hitters.shape[0]\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e302823e",
   "metadata": {},
   "source": [
    " We will fit two linear models (least squares  and lasso) and  compare their performance\n",
    "to that of a neural network. For this comparison we will use mean absolute error on a validation dataset.\n",
    "\\begin{equation*}\n",
    "\\begin{split}\n",
    "\\mbox{MAE}(y,\\hat{y}) = \\frac{1}{n} \\sum_{i=1}^n |y_i-\\hat{y}_i|.\n",
    "\\end{split}\n",
    "\\end{equation*}\n",
    "We set up the model matrix and the response."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "4deacf91",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "model = MS(Hitters.columns.drop('Salary'), intercept=False)\n",
    "X = model.fit_transform(Hitters).to_numpy()\n",
    "Y = Hitters['Salary'].to_numpy()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d030ccab",
   "metadata": {},
   "source": [
    "The `to_numpy()`  method above converts `pandas`\n",
    "data frames or series to `numpy` arrays.\n",
    "We do this because we will need to  use `sklearn` to fit the lasso model,\n",
    "and it requires this conversion. \n",
    "We also use  a linear regression method from `sklearn`, rather than the method\n",
    "in Chapter~3 from `statsmodels`, to facilitate the comparisons."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "13ffb53a",
   "metadata": {},
   "source": [
    "We now split the data into test and training, fixing the random\n",
    "state used by `sklearn` to do the split."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "8593f62b",
   "metadata": {},
   "outputs": [],
   "source": [
    "(X_train, \n",
    " X_test,\n",
    " Y_train,\n",
    " Y_test) = train_test_split(X,\n",
    "                            Y,\n",
    "                            test_size=1/3,\n",
    "                            random_state=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d1e980ad",
   "metadata": {},
   "source": [
    "### Linear Models\n",
    "We fit the linear model and evaluate the test error directly."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "54b14993",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "259.71528833146243"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "hit_lm = LinearRegression().fit(X_train, Y_train)\n",
    "Yhat_test = hit_lm.predict(X_test)\n",
    "np.abs(Yhat_test - Y_test).mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c18abe13",
   "metadata": {},
   "source": [
    "Next we fit the lasso using `sklearn`. We are using\n",
    "mean absolute error to select and evaluate a model, rather than mean squared error.\n",
    "The specialized solver we used in Section~\\ref{Ch6-varselect-lab:lab-2-ridge-regression-and-the-lasso} uses only mean squared error. So here, with a bit more work,  we create a cross-validation grid and perform the cross-validation directly.  \n",
    "\n",
    "We encode a pipeline with two steps: we first normalize the features using a `StandardScaler()` transform,\n",
    "and then fit the lasso without further normalization. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "84c668ad",
   "metadata": {},
   "outputs": [],
   "source": [
    "scaler = StandardScaler(with_mean=True, with_std=True)\n",
    "lasso = Lasso(warm_start=True, max_iter=30000)\n",
    "standard_lasso = Pipeline(steps=[('scaler', scaler),\n",
    "                                 ('lasso', lasso)])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "aab2c215",
   "metadata": {},
   "source": [
    "We need to create a grid of values for $\\lambda$. As is common practice, \n",
    "we choose a grid of 100 values of $\\lambda$, uniform on the log scale from `lam_max` down to  `0.01*lam_max`. Here  `lam_max` is the smallest value of\n",
    "$\\lambda$ with an  all-zero solution. This value equals the largest absolute inner-product between any predictor and the (centered) response. {The derivation of this result is beyond the scope of this book.}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "093f8ce9",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "X_s = scaler.fit_transform(X_train)\n",
    "n = X_s.shape[0]\n",
    "lam_max = np.fabs(X_s.T.dot(Y_train - Y_train.mean())).max() / n\n",
    "param_grid = {'lasso__alpha': np.exp(np.linspace(0, np.log(0.01), 100))\n",
    "             * lam_max}"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "def35611",
   "metadata": {},
   "source": [
    "Note that we had to transform the data first, since the scale of the variables impacts the choice of $\\lambda$.\n",
    "We now perform cross-validation using this sequence of $\\lambda$ values."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "38773381",
   "metadata": {},
   "outputs": [],
   "source": [
    "cv = KFold(10,\n",
    "           shuffle=True,\n",
    "           random_state=1)\n",
    "grid = GridSearchCV(standard_lasso,\n",
    "                    param_grid,\n",
    "                    cv=cv,\n",
    "                    scoring='neg_mean_absolute_error')\n",
    "grid.fit(X_train, Y_train);"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "53f2dc95",
   "metadata": {},
   "source": [
    "We extract the lasso model with best cross-validated mean absolute error, and evaluate its\n",
    "performance on `X_test` and `Y_test`, which were not used in\n",
    "cross-validation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "adc24a51",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "235.67548374780287"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "trained_lasso = grid.best_estimator_\n",
    "Yhat_test = trained_lasso.predict(X_test)\n",
    "np.fabs(Yhat_test - Y_test).mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "252e9dc1",
   "metadata": {},
   "source": [
    "This is similar to the results we got for the linear model fit by least squares. However, these results can vary a lot for different train/test splits; we encourage the reader to try a different seed in code block 12 and rerun the subsequent code up to this point.\n",
    "\n",
    "### Specifying a Network: Classes and Inheritance\n",
    "To fit the neural network, we first set up a model structure\n",
    "that describes the network.\n",
    "Doing so requires us to define new classes specific to the model we wish to fit.\n",
    "Typically this is done in  `pytorch` by sub-classing a generic\n",
    "representation of a network, which is the approach we take here.\n",
    "Although this example is simple, we will go through the steps in some detail, since it will serve us well\n",
    "for the more complex examples to follow.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "415cf74b",
   "metadata": {},
   "outputs": [],
   "source": [
    "class HittersModel(nn.Module):\n",
    "\n",
    "    def __init__(self, input_size):\n",
    "        super(HittersModel, self).__init__()\n",
    "        self.flatten = nn.Flatten()\n",
    "        self.sequential = nn.Sequential(\n",
    "            nn.Linear(input_size, 50),\n",
    "            nn.ReLU(),\n",
    "            nn.Dropout(0.4),\n",
    "            nn.Linear(50, 1))\n",
    "\n",
    "    def forward(self, x):\n",
    "        x = self.flatten(x)\n",
    "        return torch.flatten(self.sequential(x))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ca326fd5",
   "metadata": {},
   "source": [
    "The `class` statement identifies the code chunk as a\n",
    "declaration for a class `HittersModel`\n",
    "that inherits from the  base class `nn.Module`. This base\n",
    "class is ubiquitous in `torch` and represents the\n",
    "mappings in the neural networks.\n",
    "\n",
    "Indented beneath the `class` statement are the methods of this class:\n",
    "in this case `__init__` and `forward`.  The `__init__` method is\n",
    "called when an instance of the class is created as in the cell\n",
    "below. In the methods, `self` always refers to an instance of the\n",
    "class. In the `__init__` method, we have attached two objects to\n",
    "`self` as attributes: `flatten` and `sequential`. These are used in\n",
    "the `forward` method to describe the map that this module implements.\n",
    "\n",
    "There is one additional line in the `__init__` method, which\n",
    "is a call to\n",
    "`super()`. This function allows subclasses (i.e. `HittersModel`)\n",
    "to access methods of the class they inherit from. For example,\n",
    "the class `nn.Module` has its own `__init__` method, which is different from\n",
    "the `HittersModel.__init__()` method we’ve written above.\n",
    "Using `super()` allows us to call the method of the base class. For\n",
    "`torch` models, we will always be making this `super()` call as it is necessary\n",
    "for the model to be properly interpreted by `torch`.\n",
    "\n",
    "The object `nn.Module` has more methods than simply `__init__` and `forward`. These\n",
    "methods are directly accessible to `HittersModel` instances because of this inheritance.\n",
    "One such method we will see shortly is the `eval()` method, used\n",
    "to disable dropout for when we want to evaluate the model on test data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "60469c96",
   "metadata": {},
   "outputs": [],
   "source": [
    "hit_model = HittersModel(X.shape[1])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "114db3d4",
   "metadata": {},
   "source": [
    "The object `self.sequential` is a composition of four maps. The\n",
    "first maps the 19 features of `Hitters` to 50 dimensions, introducing $50\\times 19+50$ parameters\n",
    "for the weights and *intercept*  of the map (often called the *bias*). This layer\n",
    "is then mapped to a ReLU layer followed by a 40% dropout layer, and finally a\n",
    "linear map down to 1 dimension, again with a bias. The total number of\n",
    "trainable parameters is therefore $50\\times 19+50+50+1=1051$.\n",
    "\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9f5be00d",
   "metadata": {},
   "source": [
    "The package `torchinfo` provides a `summary()` function that neatly summarizes\n",
    "this information. We specify the size of the input and see the size\n",
    "of each tensor as it passes through layers of the network. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "84720ca7",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "===================================================================================================================\n",
       "Layer (type:depth-idx)                   Input Shape               Output Shape              Param #\n",
       "===================================================================================================================\n",
       "HittersModel                             [175, 19]                 [175]                     --\n",
       "├─Flatten: 1-1                           [175, 19]                 [175, 19]                 --\n",
       "├─Sequential: 1-2                        [175, 19]                 [175, 1]                  --\n",
       "│    └─Linear: 2-1                       [175, 19]                 [175, 50]                 1,000\n",
       "│    └─ReLU: 2-2                         [175, 50]                 [175, 50]                 --\n",
       "│    └─Dropout: 2-3                      [175, 50]                 [175, 50]                 --\n",
       "│    └─Linear: 2-4                       [175, 50]                 [175, 1]                  51\n",
       "===================================================================================================================\n",
       "Total params: 1,051\n",
       "Trainable params: 1,051\n",
       "Non-trainable params: 0\n",
       "Total mult-adds (Units.MEGABYTES): 0.18\n",
       "===================================================================================================================\n",
       "Input size (MB): 0.01\n",
       "Forward/backward pass size (MB): 0.07\n",
       "Params size (MB): 0.00\n",
       "Estimated Total Size (MB): 0.09\n",
       "==================================================================================================================="
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "summary(hit_model, \n",
    "        input_size=X_train.shape,\n",
    "        col_names=['input_size',\n",
    "                   'output_size',\n",
    "                   'num_params'])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "48833aad",
   "metadata": {},
   "source": [
    "We have truncated the end of the output slightly, here and in subsequent uses.\n",
    "\n",
    "We now need to transform our training data into a form accessible to `torch`.\n",
    "The basic\n",
    "datatype in `torch` is a `tensor`, which is very similar\n",
    "to an `ndarray` from early chapters.\n",
    "We also note here that `torch` typically\n",
    "works with 32-bit (*single precision*)\n",
    "rather than 64-bit (*double precision*) floating point numbers.\n",
    "We therefore convert our data to `np.float32` before\n",
    "forming the tensor.\n",
    "The $X$ and $Y$ tensors are then arranged into a `Dataset`\n",
    "recognized by `torch`\n",
    "using `TensorDataset()`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "6f8038ce",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "X_train_t = torch.tensor(X_train.astype(np.float32))\n",
    "Y_train_t = torch.tensor(Y_train.astype(np.float32))\n",
    "hit_train = TensorDataset(X_train_t, Y_train_t)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3b0c854a",
   "metadata": {},
   "source": [
    "We do the same for the test data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "ffe8314e",
   "metadata": {},
   "outputs": [],
   "source": [
    "X_test_t = torch.tensor(X_test.astype(np.float32))\n",
    "Y_test_t = torch.tensor(Y_test.astype(np.float32))\n",
    "hit_test = TensorDataset(X_test_t, Y_test_t)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8806a619",
   "metadata": {},
   "source": [
    "Finally, this dataset is passed to a `DataLoader()` which ultimately\n",
    "passes data into our network. While this may seem\n",
    "like a lot of overhead, this structure is helpful for more\n",
    "complex tasks where data may live on different machines,\n",
    "or where data must be passed to a GPU.\n",
    "We provide a helper function `SimpleDataModule()` in `ISLP` to make this task easier for\n",
    "standard usage.\n",
    "One of its arguments is `num_workers`, which indicates\n",
    "how many processes we will use\n",
    "for loading the data. For small\n",
    "data like `Hitters` this will have little effect, but\n",
    "it does provide an advantage for the `MNIST`  and `CIFAR100` examples below.\n",
    "The `torch` package will inspect the process running and determine a\n",
    "maximum number of workers. {This depends on the computing hardware and the number of cores available.} We’ve included a function\n",
    "`rec_num_workers()` to compute this so we know how many\n",
    "workers might be reasonable (here the max was 16)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "337f4f07",
   "metadata": {},
   "outputs": [],
   "source": [
    "max_num_workers = rec_num_workers()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "18928238",
   "metadata": {},
   "source": [
    "The general training setup in `pytorch_lightning` involves\n",
    "training, validation and test data. These are each\n",
    "represented by different data loaders. During each epoch,\n",
    "we run a training step to learn the model and a validation\n",
    "step to track the error. The test data is typically\n",
    "used at the end of training to evaluate the model.\n",
    "\n",
    "In this case, as we had split only into test and training,\n",
    "we’ll use the test data as validation data with the\n",
    "argument `validation=hit_test`. The\n",
    "`validation` argument can be a float between 0 and 1, an\n",
    "integer, or a\n",
    "`Dataset`. If a float (respectively, integer), it is interpreted\n",
    "as a percentage (respectively number) of the *training* observations to be used for validation.\n",
    "If it is a `Dataset`, it is passed directly to a data loader."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "c60b1e3f",
   "metadata": {},
   "outputs": [],
   "source": [
    "hit_dm = SimpleDataModule(hit_train,\n",
    "                          hit_test,\n",
    "                          batch_size=32,\n",
    "                          num_workers=min(4, max_num_workers),\n",
    "                          validation=hit_test)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9e5f527d",
   "metadata": {},
   "source": [
    "Next we must provide a `pytorch_lightning` module that controls\n",
    "the steps performed during the training process. We provide methods for our\n",
    "`SimpleModule()` that simply record the value\n",
    "of the loss function and any additional\n",
    "metrics at the end of each epoch. These operations\n",
    "are controlled by the methods `SimpleModule.[training/test/validation]_step()`, though\n",
    "we will not be modifying these in our examples."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "916cfd66",
   "metadata": {},
   "outputs": [],
   "source": [
    "hit_module = SimpleModule.regression(hit_model,\n",
    "                           metrics={'mae':MeanAbsoluteError()})\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cf7f2d1b",
   "metadata": {},
   "source": [
    " By using the `SimpleModule.regression()` method,  we indicate that we will use squared-error loss as in\n",
    "(\\ref{Ch13:eq:4}).\n",
    "We have also asked for mean absolute error to be tracked as well\n",
    "in the metrics that are logged.\n",
    "\n",
    "We log our results via `CSVLogger()`, which in this case stores the results in a CSV file within a directory `logs/hitters`. After the fitting is complete, this allows us to load the\n",
    "results as a `pd.DataFrame()` and visualize them below. There are\n",
    "several ways to log the results within `pytorch_lightning`, though\n",
    "we will not cover those here in detail."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "44a44fb1",
   "metadata": {},
   "outputs": [],
   "source": [
    "hit_logger = CSVLogger('logs', name='hitters')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f37f19f0",
   "metadata": {},
   "source": [
    "Finally we are ready to train our model and log the results. We\n",
    "use the `Trainer()` object from `pytorch_lightning`\n",
    "to do this work. The argument `datamodule=hit_dm` tells the trainer\n",
    "how training/validation/test logs are produced,\n",
    "while the first argument `hit_module`\n",
    "specifies the network architecture\n",
    "as well as the training/validation/test steps.\n",
    "The `callbacks` argument allows for\n",
    "several tasks to be carried out at various\n",
    "points while training a model. Here\n",
    "our `ErrorTracker()` callback will enable\n",
    "us to compute validation error while training\n",
    "and, finally, the test error.\n",
    "We now fit the model for 50 epochs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "8e91122a",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "GPU available: True (mps), used: True\n",
      "TPU available: False, using: 0 TPU cores\n",
      "IPU available: False, using: 0 IPUs\n",
      "HPU available: False, using: 0 HPUs\n",
      "\n",
      "  | Name  | Type         | Params\n",
      "---------------------------------------\n",
      "0 | model | HittersModel | 1.1 K \n",
      "1 | loss  | MSELoss      | 0     \n",
      "---------------------------------------\n",
      "1.1 K     Trainable params\n",
      "0         Non-trainable params\n",
      "1.1 K     Total params\n",
      "0.004     Total estimated model params size (MB)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 0: 100%|█████████████████████████████████████████████████████████████████████████████| 6/6 [00:01<00:00,  4.03it/s, v_num=7]\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 834.19it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 567.56it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 189.22it/s]\u001b[A\n",
      "Epoch 1: 100%|████████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 228.62it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 867.31it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 690.93it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 571.77it/s]\u001b[A\n",
      "Epoch 2: 100%|████████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 231.96it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 729.44it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 561.30it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 511.08it/s]\u001b[A\n",
      "Epoch 3: 100%|████████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 241.35it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 671.52it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 526.59it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 486.92it/s]\u001b[A\n",
      "Epoch 4: 100%|████████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 232.41it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 759.84it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 562.24it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 531.82it/s]\u001b[A\n",
      "Epoch 5: 100%|████████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 225.74it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 750.19it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 569.80it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 539.85it/s]\u001b[A\n",
      "Epoch 6: 100%|████████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 236.54it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 784.13it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 569.96it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 515.63it/s]\u001b[A\n",
      "Epoch 7: 100%|████████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 228.05it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 733.01it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 567.68it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 536.29it/s]\u001b[A\n",
      "Epoch 8: 100%|████████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 238.27it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 729.44it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 575.90it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 549.98it/s]\u001b[A\n",
      "Epoch 9: 100%|████████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 233.54it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 695.00it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 571.28it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 575.69it/s]\u001b[A\n",
      "Epoch 10: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 141.10it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 694.42it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 514.70it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 477.19it/s]\u001b[A\n",
      "Epoch 11: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 211.01it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 765.80it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 539.84it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 520.75it/s]\u001b[A\n",
      "Epoch 12: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 228.19it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 717.96it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 580.04it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 521.74it/s]\u001b[A\n",
      "Epoch 13: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 226.32it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 696.84it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 577.37it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 550.77it/s]\u001b[A\n",
      "Epoch 14: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 235.64it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 747.38it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 565.27it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 545.23it/s]\u001b[A\n",
      "Epoch 15: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 226.05it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 540.50it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 435.73it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 446.16it/s]\u001b[A\n",
      "Epoch 16: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 213.13it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 773.43it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 628.74it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 582.73it/s]\u001b[A\n",
      "Epoch 17: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 231.60it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 654.44it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 551.27it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 539.76it/s]\u001b[A\n",
      "Epoch 18: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 233.47it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 767.48it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 567.56it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 545.35it/s]\u001b[A\n",
      "Epoch 19: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 223.37it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 797.40it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 647.02it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 599.16it/s]\u001b[A\n",
      "Epoch 20: 100%|████████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 76.45it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 742.88it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 584.94it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 565.50it/s]\u001b[A\n",
      "Epoch 21: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 230.47it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 763.99it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 554.33it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 510.46it/s]\u001b[A\n",
      "Epoch 22: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 240.44it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 624.99it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 554.18it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 514.85it/s]\u001b[A\n",
      "Epoch 23: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 239.80it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 620.37it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 529.38it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 520.19it/s]\u001b[A\n",
      "Epoch 24: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 235.10it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 780.63it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 587.03it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 572.73it/s]\u001b[A\n",
      "Epoch 25: 100%|████████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 77.30it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 811.59it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 631.29it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 531.53it/s]\u001b[A\n",
      "Epoch 26: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 228.06it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 700.80it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 573.07it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 541.13it/s]\u001b[A\n",
      "Epoch 27: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 231.81it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 697.42it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 552.68it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 521.92it/s]\u001b[A\n",
      "Epoch 28: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 239.59it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 719.93it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 607.39it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 575.69it/s]\u001b[A\n",
      "Epoch 29: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 236.81it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 657.52it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 540.54it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 533.06it/s]\u001b[A\n",
      "Epoch 30: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 239.42it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 127.87it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████▋                       | 2/3 [00:00<00:00, 54.91it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|██████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 60.58it/s]\u001b[A\n",
      "Epoch 31: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 189.50it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 803.97it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 600.09it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 559.29it/s]\u001b[A\n",
      "Epoch 32: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 230.71it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 777.73it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 620.55it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 590.91it/s]\u001b[A\n",
      "Epoch 33: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 235.73it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 704.81it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 573.23it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 540.55it/s]\u001b[A\n",
      "Epoch 34: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 239.90it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 737.91it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 546.88it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 557.73it/s]\u001b[A\n",
      "Epoch 35: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 227.41it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 781.21it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 572.60it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 567.44it/s]\u001b[A\n",
      "Epoch 36: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 155.57it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 641.82it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 513.76it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 498.73it/s]\u001b[A\n",
      "Epoch 37: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 218.34it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 721.04it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 582.10it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 550.77it/s]\u001b[A\n",
      "Epoch 38: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 231.09it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 809.09it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 582.58it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 568.18it/s]\u001b[A\n",
      "Epoch 39: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 230.89it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 844.60it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 533.36it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 529.56it/s]\u001b[A\n",
      "Epoch 40: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 228.34it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 730.59it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 660.52it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 597.51it/s]\u001b[A\n",
      "Epoch 41: 100%|████████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 85.16it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 641.82it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 500.36it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 487.58it/s]\u001b[A\n",
      "Epoch 42: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 230.32it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 839.36it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 644.68it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 581.95it/s]\u001b[A\n",
      "Epoch 43: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 235.44it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 794.98it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 649.12it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 567.21it/s]\u001b[A\n",
      "Epoch 44: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 237.61it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 716.24it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 610.12it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 585.12it/s]\u001b[A\n",
      "Epoch 45: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 223.45it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 900.26it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 692.53it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 624.77it/s]\u001b[A\n",
      "Epoch 46: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 229.32it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 800.59it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 657.67it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 593.79it/s]\u001b[A\n",
      "Epoch 47: 100%|████████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 76.48it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 724.53it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 598.97it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 584.54it/s]\u001b[A\n",
      "Epoch 48: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 231.38it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 889.57it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 610.52it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 592.16it/s]\u001b[A\n",
      "Epoch 49: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 229.69it/s, v_num=7]\u001b[A\n",
      "Validation: |                                                                                               | 0/? [00:00<?, ?it/s]\u001b[A\n",
      "Validation:   0%|                                                                                           | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:   0%|                                                                              | 0/3 [00:00<?, ?it/s]\u001b[A\n",
      "Validation DataLoader 0:  33%|███████████████████████                                              | 1/3 [00:00<00:00, 749.12it/s]\u001b[A\n",
      "Validation DataLoader 0:  67%|██████████████████████████████████████████████                       | 2/3 [00:00<00:00, 569.65it/s]\u001b[A\n",
      "Validation DataLoader 0: 100%|█████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 583.54it/s]\u001b[A\n",
      "Epoch 49: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 148.84it/s, v_num=7]\u001b[A"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "`Trainer.fit` stopped: `max_epochs=50` reached.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 49: 100%|███████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 131.20it/s, v_num=7]\n"
     ]
    }
   ],
   "source": [
    "hit_trainer = Trainer(deterministic=True,\n",
    "                      max_epochs=50,\n",
    "                      log_every_n_steps=5,\n",
    "                      logger=hit_logger,\n",
    "                      callbacks=[ErrorTracker()])\n",
    "hit_trainer.fit(hit_module, datamodule=hit_dm)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "67e67f2d",
   "metadata": {},
   "source": [
    "At each step of SGD, the algorithm randomly selects 32 training observations for\n",
    "the computation of the gradient. Recall from Section~\\ref{Ch13:sec:fitt-neur-netw}\n",
    "that an epoch amounts to the number of SGD steps required to process $n$\n",
    "observations. Since the training set has\n",
    "$n=175$, and we specified a `batch_size` of 32 in the construction of  `hit_dm`, an epoch is $175/32=5.5$ SGD steps.\n",
    "\n",
    "After having fit the model, we can evaluate performance on our test\n",
    "data using the `test()` method of our trainer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "e6012d1f",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Testing DataLoader 0: 100%|████████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 318.37it/s]\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\">        Test metric        </span>┃<span style=\"font-weight: bold\">       DataLoader 0        </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">         test_loss         </span>│<span style=\"color: #800080; text-decoration-color: #800080\">      107904.6484375       </span>│\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">         test_mae          </span>│<span style=\"color: #800080; text-decoration-color: #800080\">    221.83148193359375     </span>│\n",
       "└───────────────────────────┴───────────────────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1m       Test metric       \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      DataLoader 0       \u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│\u001b[36m \u001b[0m\u001b[36m        test_loss        \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m     107904.6484375      \u001b[0m\u001b[35m \u001b[0m│\n",
       "│\u001b[36m \u001b[0m\u001b[36m        test_mae         \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m   221.83148193359375    \u001b[0m\u001b[35m \u001b[0m│\n",
       "└───────────────────────────┴───────────────────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "[{'test_loss': 107904.6484375, 'test_mae': 221.83148193359375}]"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "hit_trainer.test(hit_module, datamodule=hit_dm)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "88fcc245",
   "metadata": {},
   "source": [
    "The results of the fit have been logged into a CSV file. We can find the\n",
    "results specific to this run in the `experiment.metrics_file_path`\n",
    "attribute of our logger. Note that each time the model is fit, the logger will output\n",
    "results into a new subdirectory of our directory `logs/hitters`.\n",
    "\n",
    "We now create a plot of the MAE (mean absolute error) as a function of\n",
    "the number of epochs.\n",
    "First we retrieve the logged summaries."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "48313d02",
   "metadata": {},
   "outputs": [],
   "source": [
    "hit_results = pd.read_csv(hit_logger.experiment.metrics_file_path)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b3be9d88",
   "metadata": {},
   "source": [
    "Since we will produce similar plots in later examples, we write a\n",
    "simple generic function to produce this plot."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "4f9ccf9b",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "def summary_plot(results,\n",
    "                 ax,\n",
    "                 col='loss',\n",
    "                 valid_legend='Validation',\n",
    "                 training_legend='Training',\n",
    "                 ylabel='Loss',\n",
    "                 fontsize=20):\n",
    "    for (column,\n",
    "         color,\n",
    "         label) in zip([f'train_{col}_epoch',\n",
    "                        f'valid_{col}'],\n",
    "                       ['black',\n",
    "                        'red'],\n",
    "                       [training_legend,\n",
    "                        valid_legend]):\n",
    "        results.plot(x='epoch',\n",
    "                     y=column,\n",
    "                     label=label,\n",
    "                     marker='o',\n",
    "                     color=color,\n",
    "                     ax=ax)\n",
    "    ax.set_xlabel('Epoch')\n",
    "    ax.set_ylabel(ylabel)\n",
    "    return ax"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6b080cf1",
   "metadata": {},
   "source": [
    "We now set up our axes, and use our function to produce the MAE plot."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "3d77370a",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 600x600 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = subplots(1, 1, figsize=(6, 6))\n",
    "ax = summary_plot(hit_results,\n",
    "                  ax,\n",
    "                  col='mae',\n",
    "                  ylabel='MAE',\n",
    "                  valid_legend='Validation (=Test)')\n",
    "ax.set_ylim([0, 400])\n",
    "ax.set_xticks(np.linspace(0, 50, 11).astype(int));"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ac0b2b37",
   "metadata": {},
   "source": [
    "We can predict directly from the final model, and\n",
    "evaluate its performance on the test data.\n",
    "Before fitting, we call the `eval()` method\n",
    "of `hit_model`.\n",
    "This tells\n",
    "`torch` to effectively consider this model to be fitted, so that\n",
    "we can use it to predict on new data. For our model here,\n",
    "the biggest change is that the dropout layers will\n",
    "be turned off, i.e. no weights will be randomly\n",
    "dropped in predicting on new data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "e6a48733",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor(221.8315, grad_fn=<MeanBackward0>)"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "hit_model.eval() \n",
    "preds = hit_module(X_test_t)\n",
    "torch.abs(Y_test_t - preds).mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a24663bc",
   "metadata": {},
   "source": [
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b51594bb",
   "metadata": {},
   "source": [
    "### Cleanup\n",
    "In setting up our data module, we had initiated\n",
    "several worker processes that will remain running.\n",
    "We delete all references to the torch objects to ensure these processes\n",
    "will be killed.\n",
    " "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "78ca542d",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [],
   "source": [
    "del(Hitters,\n",
    "    hit_model, hit_dm,\n",
    "    hit_logger,\n",
    "    hit_test, hit_train,\n",
    "    X, Y,\n",
    "    X_test, X_train,\n",
    "    Y_test, Y_train,\n",
    "    X_test_t, Y_test_t,\n",
    "    hit_trainer, hit_module)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "42547131",
   "metadata": {},
   "source": [
    "## Multilayer Network on the MNIST Digit Data\n",
    "The `torchvision` package comes with a number of example datasets,\n",
    "including the `MNIST`  digit data. Our first step is to retrieve\n",
    "the training and test data sets; the `MNIST()` function within\n",
    "`torchvision.datasets` is provided for this purpose. The\n",
    "data will be downloaded the first time this function is executed, and stored in the directory `data/MNIST`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "7a6d7416",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Dataset MNIST\n",
       "    Number of datapoints: 60000\n",
       "    Root location: data\n",
       "    Split: Train\n",
       "    StandardTransform\n",
       "Transform: ToTensor()"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "(mnist_train, \n",
    " mnist_test) = [MNIST(root='data',\n",
    "                      train=train,\n",
    "                      download=True,\n",
    "                      transform=ToTensor())\n",
    "                for train in [True, False]]\n",
    "mnist_train\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "23bf4a88",
   "metadata": {},
   "source": [
    "There are 60,000 images in the training data and 10,000 in the test\n",
    "data. The images are $28\\times 28$, and stored as a matrix of pixels. We\n",
    "need to transform each one into a vector.\n",
    "\n",
    "Neural networks are somewhat sensitive to the scale of the inputs, much as ridge and\n",
    "lasso regularization are affected by scaling.  Here the inputs are eight-bit\n",
    "grayscale values between 0 and 255, so we rescale to the unit\n",
    "interval. {Note: eight bits means $2^8$, which equals 256. Since the convention\n",
    "is to start at $0$, the possible values  range from $0$ to $255$.}\n",
    "This transformation, along with some reordering\n",
    "of the axes, is performed by the `ToTensor()` transform\n",
    "from the `torchvision.transforms` package.\n",
    "\n",
    "As in our `Hitters` example, we form a data module\n",
    "from the training and test datasets, setting aside 20%\n",
    "of the training images for validation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "e6c5e82c",
   "metadata": {},
   "outputs": [],
   "source": [
    "mnist_dm = SimpleDataModule(mnist_train,\n",
    "                            mnist_test,\n",
    "                            validation=0.2,\n",
    "                            num_workers=max_num_workers,\n",
    "                            batch_size=256)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "73614d27",
   "metadata": {},
   "source": [
    "Let’s take a look at the data that will get fed into our network. We loop through the first few\n",
    "chunks of the test dataset, breaking after 2 batches:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "6ecf940c",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X:  torch.Size([256, 1, 28, 28])\n",
      "Y:  torch.Size([256])\n",
      "X:  torch.Size([256, 1, 28, 28])\n",
      "Y:  torch.Size([256])\n"
     ]
    }
   ],
   "source": [
    "for idx, (X_ ,Y_) in enumerate(mnist_dm.train_dataloader()):\n",
    "    print('X: ', X_.shape)\n",
    "    print('Y: ', Y_.shape)\n",
    "    if idx >= 1:\n",
    "        break\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6e0b2d96",
   "metadata": {},
   "source": [
    "We see that the $X$ for each batch consists of 256 images of size `1x28x28`.\n",
    "Here the `1` indicates a single channel (greyscale). For RGB images such as `CIFAR100` below,\n",
    "we will see that the `1` in the size will be replaced by `3` for the three RGB channels.\n",
    "\n",
    "Now we are ready to specify our neural network."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "f4c49bab",
   "metadata": {},
   "outputs": [],
   "source": [
    "class MNISTModel(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(MNISTModel, self).__init__()\n",
    "        self.layer1 = nn.Sequential(\n",
    "            nn.Flatten(),\n",
    "            nn.Linear(28*28, 256),\n",
    "            nn.ReLU(),\n",
    "            nn.Dropout(0.4))\n",
    "        self.layer2 = nn.Sequential(\n",
    "            nn.Linear(256, 128),\n",
    "            nn.ReLU(),\n",
    "            nn.Dropout(0.3))\n",
    "        self._forward = nn.Sequential(\n",
    "            self.layer1,\n",
    "            self.layer2,\n",
    "            nn.Linear(128, 10))\n",
    "    def forward(self, x):\n",
    "        return self._forward(x)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7e25c568",
   "metadata": {},
   "source": [
    "We see that in the first layer, each `1x28x28` image is flattened, then mapped to\n",
    "256 dimensions where we apply a ReLU activation with 40% dropout.\n",
    "A second layer maps the first layer’s output down to\n",
    "128 dimensions, applying a ReLU activation with 30% dropout. Finally,\n",
    "the 128 dimensions are mapped down to 10, the number of classes in the\n",
    "`MNIST`  data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "8dd2770f",
   "metadata": {},
   "outputs": [],
   "source": [
    "mnist_model = MNISTModel()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3d0e8c1b",
   "metadata": {},
   "source": [
    "We can check that the model produces output of expected size based\n",
    "on our existing batch `X_` above."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "c916312b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([256, 10])"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mnist_model(X_).size()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a3158809",
   "metadata": {},
   "source": [
    "Let’s take a look at the summary of the model. Instead of an `input_size` we can pass\n",
    "a tensor of correct shape. In this case, we pass through the final\n",
    "batched `X_` from above."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "54a6b6b2",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "===================================================================================================================\n",
       "Layer (type:depth-idx)                   Input Shape               Output Shape              Param #\n",
       "===================================================================================================================\n",
       "MNISTModel                               [256, 1, 28, 28]          [256, 10]                 --\n",
       "├─Sequential: 1-1                        [256, 1, 28, 28]          [256, 10]                 --\n",
       "│    └─Sequential: 2-1                   [256, 1, 28, 28]          [256, 256]                --\n",
       "│    │    └─Flatten: 3-1                 [256, 1, 28, 28]          [256, 784]                --\n",
       "│    │    └─Linear: 3-2                  [256, 784]                [256, 256]                200,960\n",
       "│    │    └─ReLU: 3-3                    [256, 256]                [256, 256]                --\n",
       "│    │    └─Dropout: 3-4                 [256, 256]                [256, 256]                --\n",
       "│    └─Sequential: 2-2                   [256, 256]                [256, 128]                --\n",
       "│    │    └─Linear: 3-5                  [256, 256]                [256, 128]                32,896\n",
       "│    │    └─ReLU: 3-6                    [256, 128]                [256, 128]                --\n",
       "│    │    └─Dropout: 3-7                 [256, 128]                [256, 128]                --\n",
       "│    └─Linear: 2-3                       [256, 128]                [256, 10]                 1,290\n",
       "===================================================================================================================\n",
       "Total params: 235,146\n",
       "Trainable params: 235,146\n",
       "Non-trainable params: 0\n",
       "Total mult-adds (Units.MEGABYTES): 60.20\n",
       "===================================================================================================================\n",
       "Input size (MB): 0.80\n",
       "Forward/backward pass size (MB): 0.81\n",
       "Params size (MB): 0.94\n",
       "Estimated Total Size (MB): 2.55\n",
       "==================================================================================================================="
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "summary(mnist_model,\n",
    "        input_data=X_,\n",
    "        col_names=['input_size',\n",
    "                   'output_size',\n",
    "                   'num_params'])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e76a07df",
   "metadata": {},
   "source": [
    "Having set up both  the model and the data module, fitting this model is\n",
    "now almost identical to the `Hitters` example. In contrast to our regression model, here we will use the\n",
    "`SimpleModule.classification()` method which\n",
    "uses the  cross-entropy loss function instead of mean squared error. It must be supplied with the number of\n",
    "classes in the problem."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "c1fa96fd",
   "metadata": {},
   "outputs": [],
   "source": [
    "mnist_module = SimpleModule.classification(mnist_model,\n",
    "                                           num_classes=10)\n",
    "mnist_logger = CSVLogger('logs', name='MNIST')\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "106c0f38",
   "metadata": {},
   "source": [
    "Now we are ready to go. The final step is to supply training data, and fit the model. We disable the progress bar below to avoid lengthy output in the browser when running."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "25515af7",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "GPU available: True (mps), used: True\n",
      "TPU available: False, using: 0 TPU cores\n",
      "IPU available: False, using: 0 IPUs\n",
      "HPU available: False, using: 0 HPUs\n",
      "\n",
      "  | Name  | Type             | Params\n",
      "-------------------------------------------\n",
      "0 | model | MNISTModel       | 235 K \n",
      "1 | loss  | CrossEntropyLoss | 0     \n",
      "-------------------------------------------\n",
      "235 K     Trainable params\n",
      "0         Non-trainable params\n",
      "235 K     Total params\n",
      "0.941     Total estimated model params size (MB)\n",
      "`Trainer.fit` stopped: `max_epochs=30` reached.\n"
     ]
    }
   ],
   "source": [
    "mnist_trainer = Trainer(deterministic=True,\n",
    "                        max_epochs=30,\n",
    "                        logger=mnist_logger,\n",
    "                        enable_progress_bar=False,\n",
    "                        callbacks=[ErrorTracker()])\n",
    "mnist_trainer.fit(mnist_module,\n",
    "                  datamodule=mnist_dm)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0e12d8c7",
   "metadata": {},
   "source": [
    "We have suppressed the output here, which is a progress report on the\n",
    "fitting of the model, grouped by epoch. This is very useful, since on\n",
    "large datasets fitting can take time. Fitting this model took 245\n",
    "seconds on a MacBook Pro with an Apple M1 Pro chip with 10 cores and 16 GB of RAM.\n",
    "Here we specified a\n",
    "validation split of 20%, so training is actually performed on\n",
    "80% of the 60,000 observations in the training set. This is an\n",
    "alternative to actually supplying validation data, like we did for the `Hitters` data.\n",
    "SGD  uses batches\n",
    "of 256 observations in computing the gradient, and doing the\n",
    "arithmetic, we see that an epoch corresponds to 188 gradient steps."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "eb25107e",
   "metadata": {},
   "source": [
    "`SimpleModule.classification()` includes\n",
    "an accuracy metric by default. Other\n",
    "classification metrics can be added from `torchmetrics`.\n",
    "We will use  our `summary_plot()` function to display \n",
    "accuracy across epochs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "872fed35",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 600x600 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "mnist_results = pd.read_csv(mnist_logger.experiment.metrics_file_path)\n",
    "fig, ax = subplots(1, 1, figsize=(6, 6))\n",
    "summary_plot(mnist_results,\n",
    "             ax,\n",
    "             col='accuracy',\n",
    "             ylabel='Accuracy')\n",
    "ax.set_ylim([0.5, 1])\n",
    "ax.set_ylabel('Accuracy')\n",
    "ax.set_xticks(np.linspace(0, 30, 7).astype(int));\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5e1c8d5c",
   "metadata": {},
   "source": [
    "Once again we evaluate the accuracy using the `test()` method of our trainer. This model achieves\n",
    "97% accuracy on the test data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "4551c956",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\">        Test metric        </span>┃<span style=\"font-weight: bold\">       DataLoader 0        </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">       test_accuracy       </span>│<span style=\"color: #800080; text-decoration-color: #800080\">    0.9620000123977661     </span>│\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">         test_loss         </span>│<span style=\"color: #800080; text-decoration-color: #800080\">    0.15120187401771545    </span>│\n",
       "└───────────────────────────┴───────────────────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1m       Test metric       \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      DataLoader 0       \u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│\u001b[36m \u001b[0m\u001b[36m      test_accuracy      \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m   0.9620000123977661    \u001b[0m\u001b[35m \u001b[0m│\n",
       "│\u001b[36m \u001b[0m\u001b[36m        test_loss        \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m   0.15120187401771545   \u001b[0m\u001b[35m \u001b[0m│\n",
       "└───────────────────────────┴───────────────────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "[{'test_loss': 0.15120187401771545, 'test_accuracy': 0.9620000123977661}]"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mnist_trainer.test(mnist_module,\n",
    "                   datamodule=mnist_dm)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1b0c0bd0",
   "metadata": {},
   "source": [
    "Table~\\ref{Ch13:tab:mnist} also reports the error rates resulting from LDA (Chapter~\\ref{Ch4:classification}) and multiclass logistic\n",
    "regression. For LDA we refer the reader to Section~\\ref{Ch4-classification-lab:linear-discriminant-analysis}.\n",
    "Although we could use the `sklearn` function `LogisticRegression()` to fit  \n",
    "multiclass logistic regression, we are set up here to fit such a model\n",
    "with `torch`.\n",
    "We just have an input layer and an output layer, and omit the hidden layers!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "id": "cbf4d0d2",
   "metadata": {},
   "outputs": [],
   "source": [
    "class MNIST_MLR(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(MNIST_MLR, self).__init__()\n",
    "        self.linear = nn.Sequential(nn.Flatten(),\n",
    "                                    nn.Linear(784, 10))\n",
    "    def forward(self, x):\n",
    "        return self.linear(x)\n",
    "\n",
    "mlr_model = MNIST_MLR()\n",
    "mlr_module = SimpleModule.classification(mlr_model,\n",
    "                                         num_classes=10)\n",
    "mlr_logger = CSVLogger('logs', name='MNIST_MLR')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "id": "6ff23014",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "GPU available: True (mps), used: True\n",
      "TPU available: False, using: 0 TPU cores\n",
      "IPU available: False, using: 0 IPUs\n",
      "HPU available: False, using: 0 HPUs\n",
      "/Users/jtaylo/anaconda3/envs/ISLP_v22_312/lib/python3.12/site-packages/pytorch_lightning/trainer/connectors/logger_connector/logger_connector.py:75: Starting from v1.9.0, `tensorboardX` has been removed as a dependency of the `pytorch_lightning` package, due to potential conflicts with other packages in the ML ecosystem. For this reason, `logger=True` will use `CSVLogger` as the default logger, unless the `tensorboard` or `tensorboardX` packages are found. Please `pip install lightning[extra]` or one of them to enable TensorBoard support by default\n",
      "\n",
      "  | Name  | Type             | Params\n",
      "-------------------------------------------\n",
      "0 | model | MNIST_MLR        | 7.9 K \n",
      "1 | loss  | CrossEntropyLoss | 0     \n",
      "-------------------------------------------\n",
      "7.9 K     Trainable params\n",
      "0         Non-trainable params\n",
      "7.9 K     Total params\n",
      "0.031     Total estimated model params size (MB)\n",
      "`Trainer.fit` stopped: `max_epochs=30` reached.\n"
     ]
    }
   ],
   "source": [
    "mlr_trainer = Trainer(deterministic=True,\n",
    "                      max_epochs=30,\n",
    "                      enable_progress_bar=False,\n",
    "                      callbacks=[ErrorTracker()])\n",
    "mlr_trainer.fit(mlr_module, datamodule=mnist_dm)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f8368ac1",
   "metadata": {},
   "source": [
    "We fit the model just as before and compute the test results."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "id": "ca159ef2",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\">        Test metric        </span>┃<span style=\"font-weight: bold\">       DataLoader 0        </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">       test_accuracy       </span>│<span style=\"color: #800080; text-decoration-color: #800080\">     0.916100025177002     </span>│\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">         test_loss         </span>│<span style=\"color: #800080; text-decoration-color: #800080\">    0.3469300866127014     </span>│\n",
       "└───────────────────────────┴───────────────────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1m       Test metric       \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      DataLoader 0       \u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│\u001b[36m \u001b[0m\u001b[36m      test_accuracy      \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m    0.916100025177002    \u001b[0m\u001b[35m \u001b[0m│\n",
       "│\u001b[36m \u001b[0m\u001b[36m        test_loss        \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m   0.3469300866127014    \u001b[0m\u001b[35m \u001b[0m│\n",
       "└───────────────────────────┴───────────────────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "[{'test_loss': 0.3469300866127014, 'test_accuracy': 0.916100025177002}]"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mlr_trainer.test(mlr_module,\n",
    "                 datamodule=mnist_dm)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2be94ca6",
   "metadata": {},
   "source": [
    "The accuracy is above 90% even for this pretty simple model.\n",
    "\n",
    "As in the `Hitters` example, we delete some of\n",
    "the objects we created above."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "5842c930",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [],
   "source": [
    "del(mnist_test,\n",
    "    mnist_train,\n",
    "    mnist_model,\n",
    "    mnist_dm,\n",
    "    mnist_trainer,\n",
    "    mnist_module,\n",
    "    mnist_results,\n",
    "    mlr_model,\n",
    "    mlr_module,\n",
    "    mlr_trainer)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cd9e98b5",
   "metadata": {},
   "source": [
    "## Convolutional Neural Networks\n",
    "In this section we fit a CNN to the `CIFAR100` data, which is available in the `torchvision`\n",
    "package. It is arranged in a similar fashion as the `MNIST` data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "cb406355",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Files already downloaded and verified\n",
      "Files already downloaded and verified\n"
     ]
    }
   ],
   "source": [
    "(cifar_train, \n",
    " cifar_test) = [CIFAR100(root=\"data\",\n",
    "                         train=train,\n",
    "                         download=True)\n",
    "             for train in [True, False]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "8dcd79cc",
   "metadata": {},
   "outputs": [],
   "source": [
    "transform = ToTensor()\n",
    "cifar_train_X = torch.stack([transform(x) for x in\n",
    "                            cifar_train.data])\n",
    "cifar_test_X = torch.stack([transform(x) for x in\n",
    "                            cifar_test.data])\n",
    "cifar_train = TensorDataset(cifar_train_X,\n",
    "                            torch.tensor(cifar_train.targets))\n",
    "cifar_test = TensorDataset(cifar_test_X,\n",
    "                            torch.tensor(cifar_test.targets))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "81f4ea20",
   "metadata": {},
   "source": [
    "The `CIFAR100` dataset consists of 50,000 training images, each represented by a three-dimensional tensor:\n",
    "each three-color image is represented as a set of three channels, each of which consists of\n",
    "$32\\times 32$ eight-bit pixels. We standardize as we did for the\n",
    "digits, but keep the array structure. This is accomplished with the `ToTensor()` transform.\n",
    "\n",
    "Creating the data module is similar to the `MNIST`  example."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "id": "6980303a",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "cifar_dm = SimpleDataModule(cifar_train,\n",
    "                            cifar_test,\n",
    "                            validation=0.2,\n",
    "                            num_workers=max_num_workers,\n",
    "                            batch_size=128)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6b9e007c",
   "metadata": {},
   "source": [
    "We again look at the shape of typical batches in our data loaders."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "id": "401888c7",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X:  torch.Size([128, 3, 32, 32])\n",
      "Y:  torch.Size([128])\n",
      "X:  torch.Size([128, 3, 32, 32])\n",
      "Y:  torch.Size([128])\n"
     ]
    }
   ],
   "source": [
    "for idx, (X_ ,Y_) in enumerate(cifar_dm.train_dataloader()):\n",
    "    print('X: ', X_.shape)\n",
    "    print('Y: ', Y_.shape)\n",
    "    if idx >= 1:\n",
    "        break\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "82438c30",
   "metadata": {},
   "source": [
    "Before we start, we look at some of the training images; similar code produced\n",
    "Figure~\\ref{Ch13:fig:cifar100} on page \\pageref{Ch13:fig:cifar100}. The example below also illustrates\n",
    "that `TensorDataset` objects can be indexed with integers --- we are choosing\n",
    "random images from the training data by indexing `cifar_train`. In order to display correctly,\n",
    "we must reorder the dimensions by a call to `np.transpose()`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "id": "56090846",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1000x1000 with 25 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, axes = subplots(5, 5, figsize=(10,10))\n",
    "rng = np.random.default_rng(4)\n",
    "indices = rng.choice(np.arange(len(cifar_train)), 25,\n",
    "                     replace=False).reshape((5,5))\n",
    "for i in range(5):\n",
    "    for j in range(5):\n",
    "        idx = indices[i,j]\n",
    "        axes[i,j].imshow(np.transpose(cifar_train[idx][0],\n",
    "                                      [1,2,0]),\n",
    "                                      interpolation=None)\n",
    "        axes[i,j].set_xticks([])\n",
    "        axes[i,j].set_yticks([])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "333dca34",
   "metadata": {},
   "source": [
    "Here the `imshow()` method recognizes from the shape of its argument that it is a 3-dimensional array, with the last dimension indexing the three RGB color channels.\n",
    "\n",
    "We specify a moderately-sized  CNN for\n",
    "demonstration purposes, similar in structure to Figure~\\ref{Ch13:fig:DeepCNN}.\n",
    "We use several layers, each consisting of  convolution, ReLU, and max-pooling steps.\n",
    "We first define a module that defines one of these layers. As in our\n",
    "previous examples, we overwrite the `__init__()` and `forward()` methods\n",
    "of `nn.Module`. This user-defined  module can now be used in ways just like\n",
    "`nn.Linear()` or `nn.Dropout()`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "id": "8c0567d9",
   "metadata": {},
   "outputs": [],
   "source": [
    "class BuildingBlock(nn.Module):\n",
    "\n",
    "    def __init__(self,\n",
    "                 in_channels,\n",
    "                 out_channels):\n",
    "\n",
    "        super(BuildingBlock, self).__init__()\n",
    "        self.conv = nn.Conv2d(in_channels=in_channels,\n",
    "                              out_channels=out_channels,\n",
    "                              kernel_size=(3,3),\n",
    "                              padding='same')\n",
    "        self.activation = nn.ReLU()\n",
    "        self.pool = nn.MaxPool2d(kernel_size=(2,2))\n",
    "\n",
    "    def forward(self, x):\n",
    "        return self.pool(self.activation(self.conv(x)))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "25956312",
   "metadata": {},
   "source": [
    "Notice that we used the `padding = \"same\"` argument to\n",
    "`nn.Conv2d()`, which ensures that the output channels have the\n",
    "same dimension as the input channels. There are 32 channels in the first\n",
    "hidden layer, in contrast to the three channels in the input layer. We\n",
    "use a $3\\times 3$ convolution filter for each channel in all the layers. Each\n",
    "convolution is followed by a max-pooling layer over $2\\times2$\n",
    "blocks.\n",
    "\n",
    "In forming our deep learning model for the `CIFAR100` data, we use several of our `BuildingBlock()`\n",
    "modules sequentially. This simple example\n",
    "illustrates some of the power of `torch`. Users can\n",
    "define modules of their own, which can be combined in other\n",
    "modules. Ultimately, everything is fit by a generic trainer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "id": "e4d3eb9a",
   "metadata": {},
   "outputs": [],
   "source": [
    "class CIFARModel(nn.Module):\n",
    "\n",
    "    def __init__(self):\n",
    "        super(CIFARModel, self).__init__()\n",
    "        sizes = [(3,32),\n",
    "                 (32,64),\n",
    "                 (64,128),\n",
    "                 (128,256)]\n",
    "        self.conv = nn.Sequential(*[BuildingBlock(in_, out_)\n",
    "                                    for in_, out_ in sizes])\n",
    "\n",
    "        self.output = nn.Sequential(nn.Dropout(0.5),\n",
    "                                    nn.Linear(2*2*256, 512),\n",
    "                                    nn.ReLU(),\n",
    "                                    nn.Linear(512, 100))\n",
    "    def forward(self, x):\n",
    "        val = self.conv(x)\n",
    "        val = torch.flatten(val, start_dim=1)\n",
    "        return self.output(val)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2cf81883",
   "metadata": {},
   "source": [
    "We  build the model and look at the summary. (We had created examples of `X_` earlier.)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "id": "5ff4e0a5",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "===================================================================================================================\n",
       "Layer (type:depth-idx)                   Input Shape               Output Shape              Param #\n",
       "===================================================================================================================\n",
       "CIFARModel                               [128, 3, 32, 32]          [128, 100]                --\n",
       "├─Sequential: 1-1                        [128, 3, 32, 32]          [128, 256, 2, 2]          --\n",
       "│    └─BuildingBlock: 2-1                [128, 3, 32, 32]          [128, 32, 16, 16]         --\n",
       "│    │    └─Conv2d: 3-1                  [128, 3, 32, 32]          [128, 32, 32, 32]         896\n",
       "│    │    └─ReLU: 3-2                    [128, 32, 32, 32]         [128, 32, 32, 32]         --\n",
       "│    │    └─MaxPool2d: 3-3               [128, 32, 32, 32]         [128, 32, 16, 16]         --\n",
       "│    └─BuildingBlock: 2-2                [128, 32, 16, 16]         [128, 64, 8, 8]           --\n",
       "│    │    └─Conv2d: 3-4                  [128, 32, 16, 16]         [128, 64, 16, 16]         18,496\n",
       "│    │    └─ReLU: 3-5                    [128, 64, 16, 16]         [128, 64, 16, 16]         --\n",
       "│    │    └─MaxPool2d: 3-6               [128, 64, 16, 16]         [128, 64, 8, 8]           --\n",
       "│    └─BuildingBlock: 2-3                [128, 64, 8, 8]           [128, 128, 4, 4]          --\n",
       "│    │    └─Conv2d: 3-7                  [128, 64, 8, 8]           [128, 128, 8, 8]          73,856\n",
       "│    │    └─ReLU: 3-8                    [128, 128, 8, 8]          [128, 128, 8, 8]          --\n",
       "│    │    └─MaxPool2d: 3-9               [128, 128, 8, 8]          [128, 128, 4, 4]          --\n",
       "│    └─BuildingBlock: 2-4                [128, 128, 4, 4]          [128, 256, 2, 2]          --\n",
       "│    │    └─Conv2d: 3-10                 [128, 128, 4, 4]          [128, 256, 4, 4]          295,168\n",
       "│    │    └─ReLU: 3-11                   [128, 256, 4, 4]          [128, 256, 4, 4]          --\n",
       "│    │    └─MaxPool2d: 3-12              [128, 256, 4, 4]          [128, 256, 2, 2]          --\n",
       "├─Sequential: 1-2                        [128, 1024]               [128, 100]                --\n",
       "│    └─Dropout: 2-5                      [128, 1024]               [128, 1024]               --\n",
       "│    └─Linear: 2-6                       [128, 1024]               [128, 512]                524,800\n",
       "│    └─ReLU: 2-7                         [128, 512]                [128, 512]                --\n",
       "│    └─Linear: 2-8                       [128, 512]                [128, 100]                51,300\n",
       "===================================================================================================================\n",
       "Total params: 964,516\n",
       "Trainable params: 964,516\n",
       "Non-trainable params: 0\n",
       "Total mult-adds (Units.GIGABYTES): 2.01\n",
       "===================================================================================================================\n",
       "Input size (MB): 1.57\n",
       "Forward/backward pass size (MB): 63.54\n",
       "Params size (MB): 3.86\n",
       "Estimated Total Size (MB): 68.97\n",
       "==================================================================================================================="
      ]
     },
     "execution_count": 56,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cifar_model = CIFARModel()\n",
    "summary(cifar_model,\n",
    "        input_data=X_,\n",
    "        col_names=['input_size',\n",
    "                   'output_size',\n",
    "                   'num_params'])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2125ae4a",
   "metadata": {},
   "source": [
    "The total number of trainable parameters is 964,516.\n",
    "By studying the size of the parameters, we can see that the channels halve in both\n",
    "dimensions\n",
    "after each of these max-pooling operations. After the last of these we\n",
    "have a layer with  256 channels of dimension $2\\times 2$. These are then\n",
    "flattened to a dense layer of size 1,024;\n",
    "in other words, each of the $2\\times 2$ matrices is turned into a\n",
    "$4$-vector, and put side-by-side in one layer. This is followed by a\n",
    "dropout regularization layer,  then\n",
    "another dense layer of size 512, and finally, the\n",
    "output layer.\n",
    "\n",
    "Up to now, we have been using a default\n",
    "optimizer in `SimpleModule()`. For these data,\n",
    "experiments show that a smaller learning rate performs\n",
    "better than the default 0.01. We use a\n",
    "custom optimizer here with a learning rate of 0.001.\n",
    "Besides this, the logging and training\n",
    "follow a similar pattern to our previous examples. The optimizer\n",
    "takes an argument `params` that informs\n",
    "the optimizer which parameters are involved in SGD (stochastic gradient descent).\n",
    "\n",
    "We saw earlier that entries of a module’s parameters are tensors. In passing\n",
    "the parameters to the optimizer we are doing more than\n",
    "simply passing arrays; part of the structure of the graph\n",
    "is encoded in the tensors themselves."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "id": "ab9d9080",
   "metadata": {},
   "outputs": [],
   "source": [
    "cifar_optimizer = RMSprop(cifar_model.parameters(), lr=0.001)\n",
    "cifar_module = SimpleModule.classification(cifar_model,\n",
    "                                    num_classes=100,\n",
    "                                    optimizer=cifar_optimizer)\n",
    "cifar_logger = CSVLogger('logs', name='CIFAR100')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "id": "5a86b37d",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "GPU available: True (mps), used: True\n",
      "TPU available: False, using: 0 TPU cores\n",
      "IPU available: False, using: 0 IPUs\n",
      "HPU available: False, using: 0 HPUs\n",
      "\n",
      "  | Name  | Type             | Params\n",
      "-------------------------------------------\n",
      "0 | model | CIFARModel       | 964 K \n",
      "1 | loss  | CrossEntropyLoss | 0     \n",
      "-------------------------------------------\n",
      "964 K     Trainable params\n",
      "0         Non-trainable params\n",
      "964 K     Total params\n",
      "3.858     Total estimated model params size (MB)\n",
      "`Trainer.fit` stopped: `max_epochs=30` reached.\n"
     ]
    }
   ],
   "source": [
    "cifar_trainer = Trainer(deterministic=True,\n",
    "                        max_epochs=30,\n",
    "                        logger=cifar_logger,\n",
    "                        enable_progress_bar=False,\n",
    "                        callbacks=[ErrorTracker()])\n",
    "cifar_trainer.fit(cifar_module,\n",
    "                  datamodule=cifar_dm)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "00c6668f",
   "metadata": {},
   "source": [
    "This model can take 10 minutes or more to run and achieves about 42% accuracy on the test\n",
    "data. Although this is not terrible for 100-class data (a random\n",
    "classifier gets 1% accuracy), searching the web we see results around\n",
    "75%. Typically it takes a lot of architecture carpentry,\n",
    "fiddling with regularization, and time, to achieve such results.\n",
    "\n",
    "Let’s take a look at the validation and training accuracy\n",
    "across epochs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "id": "9cf596b4",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 600x600 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "log_path = cifar_logger.experiment.metrics_file_path\n",
    "cifar_results = pd.read_csv(log_path)\n",
    "fig, ax = subplots(1, 1, figsize=(6, 6))\n",
    "summary_plot(cifar_results,\n",
    "             ax,\n",
    "             col='accuracy',\n",
    "             ylabel='Accuracy')\n",
    "ax.set_xticks(np.linspace(0, 10, 6).astype(int))\n",
    "ax.set_ylabel('Accuracy')\n",
    "ax.set_ylim([0, 1]);"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c3e4a8bb",
   "metadata": {},
   "source": [
    "Finally, we evaluate our model on our test data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "id": "0f5ad9bb",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\">        Test metric        </span>┃<span style=\"font-weight: bold\">       DataLoader 0        </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">       test_accuracy       </span>│<span style=\"color: #800080; text-decoration-color: #800080\">    0.4269999861717224     </span>│\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">         test_loss         </span>│<span style=\"color: #800080; text-decoration-color: #800080\">     2.45865797996521      </span>│\n",
       "└───────────────────────────┴───────────────────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1m       Test metric       \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      DataLoader 0       \u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│\u001b[36m \u001b[0m\u001b[36m      test_accuracy      \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m   0.4269999861717224    \u001b[0m\u001b[35m \u001b[0m│\n",
       "│\u001b[36m \u001b[0m\u001b[36m        test_loss        \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m    2.45865797996521     \u001b[0m\u001b[35m \u001b[0m│\n",
       "└───────────────────────────┴───────────────────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "[{'test_loss': 2.45865797996521, 'test_accuracy': 0.4269999861717224}]"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cifar_trainer.test(cifar_module,\n",
    "                   datamodule=cifar_dm)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ed52c614",
   "metadata": {},
   "source": [
    "### Hardware Acceleration\n",
    "As deep learning has become ubiquitous in machine learning, hardware\n",
    "manufacturers have produced special libraries that can\n",
    "often speed up the gradient-descent steps.\n",
    "\n",
    "For instance, Mac OS devices with the M1 chip may have the *Metal* programming framework\n",
    "enabled, which can speed up the  `torch` \n",
    "computations. We present an example of how to use this acceleration.\n",
    "\n",
    "The main changes are to the `Trainer()` call as well as to the metrics\n",
    "that will be evaluated on the data. These metrics must be told  where\n",
    "the data will be located at evaluation time. This is\n",
    "accomplished with a call to the `to()` method of the metrics."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "id": "e763b4c4",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "GPU available: True (mps), used: True\n",
      "TPU available: False, using: 0 TPU cores\n",
      "IPU available: False, using: 0 IPUs\n",
      "HPU available: False, using: 0 HPUs\n",
      "\n",
      "  | Name  | Type             | Params\n",
      "-------------------------------------------\n",
      "0 | model | CIFARModel       | 964 K \n",
      "1 | loss  | CrossEntropyLoss | 0     \n",
      "-------------------------------------------\n",
      "964 K     Trainable params\n",
      "0         Non-trainable params\n",
      "964 K     Total params\n",
      "3.858     Total estimated model params size (MB)\n",
      "`Trainer.fit` stopped: `max_epochs=30` reached.\n"
     ]
    }
   ],
   "source": [
    "try:\n",
    "    for name, metric in cifar_module.metrics.items():\n",
    "        cifar_module.metrics[name] = metric.to('mps')\n",
    "    cifar_trainer_mps = Trainer(accelerator='mps',\n",
    "                                deterministic=True,\n",
    "                                enable_progress_bar=False,\n",
    "                                max_epochs=30)\n",
    "    cifar_trainer_mps.fit(cifar_module,\n",
    "                          datamodule=cifar_dm)\n",
    "    cifar_trainer_mps.test(cifar_module,\n",
    "                          datamodule=cifar_dm)\n",
    "except:\n",
    "    pass"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "14791a8f",
   "metadata": {},
   "source": [
    "This yields approximately two- or three-fold  acceleration for each epoch.\n",
    "We have protected this code block using `try:` and `except:`\n",
    "clauses; if it works, we get the speedup, if it fails, nothing happens."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "51532403",
   "metadata": {},
   "source": [
    "## Using Pretrained CNN Models\n",
    "We now show how to use a CNN pretrained on the  `imagenet` database to classify natural\n",
    "images, and demonstrate how we produced Figure~\\ref{Ch13:fig:homeimages}.\n",
    "We copied six JPEG images from a digital photo album into the\n",
    "directory `book_images`. These images are available\n",
    "from the data section of  <www.statlearning.com>, the ISLP book website. Download `book_images.zip`; when\n",
    "clicked it creates the `book_images` directory. \n",
    "\n",
    "The pretrained network we use is called `resnet50`; specification details can be found on the web.\n",
    "We will read in the images, and\n",
    "convert them into the array format expected by the `torch`\n",
    "software to match the specifications in `resnet50`. \n",
    "The conversion involves a resize, a crop and then a predefined standardization for each of the three channels.\n",
    "We now read in the images and preprocess them."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "id": "8ddd3831",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([6, 3, 224, 224])"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "resize = Resize((232,232), antialias=True)\n",
    "crop = CenterCrop(224)\n",
    "normalize = Normalize([0.485,0.456,0.406],\n",
    "                      [0.229,0.224,0.225])\n",
    "imgfiles = sorted([f for f in glob('book_images/*')])\n",
    "imgs = torch.stack([torch.div(crop(resize(read_image(f))), 255)\n",
    "                    for f in imgfiles])\n",
    "imgs = normalize(imgs)\n",
    "imgs.size()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a735ef73",
   "metadata": {},
   "source": [
    "We now set up the trained network with the weights we read in code block~6. The model has 50 layers, with a fair bit of complexity."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "id": "1126ee25",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "===================================================================================================================\n",
       "Layer (type:depth-idx)                   Input Shape               Output Shape              Param #\n",
       "===================================================================================================================\n",
       "ResNet                                   [6, 3, 224, 224]          [6, 1000]                 --\n",
       "├─Conv2d: 1-1                            [6, 3, 224, 224]          [6, 64, 112, 112]         9,408\n",
       "├─BatchNorm2d: 1-2                       [6, 64, 112, 112]         [6, 64, 112, 112]         128\n",
       "├─ReLU: 1-3                              [6, 64, 112, 112]         [6, 64, 112, 112]         --\n",
       "├─MaxPool2d: 1-4                         [6, 64, 112, 112]         [6, 64, 56, 56]           --\n",
       "├─Sequential: 1-5                        [6, 64, 56, 56]           [6, 256, 56, 56]          --\n",
       "│    └─Bottleneck: 2-1                   [6, 64, 56, 56]           [6, 256, 56, 56]          --\n",
       "│    │    └─Conv2d: 3-1                  [6, 64, 56, 56]           [6, 64, 56, 56]           4,096\n",
       "│    │    └─BatchNorm2d: 3-2             [6, 64, 56, 56]           [6, 64, 56, 56]           128\n",
       "│    │    └─ReLU: 3-3                    [6, 64, 56, 56]           [6, 64, 56, 56]           --\n",
       "│    │    └─Conv2d: 3-4                  [6, 64, 56, 56]           [6, 64, 56, 56]           36,864\n",
       "│    │    └─BatchNorm2d: 3-5             [6, 64, 56, 56]           [6, 64, 56, 56]           128\n",
       "│    │    └─ReLU: 3-6                    [6, 64, 56, 56]           [6, 64, 56, 56]           --\n",
       "│    │    └─Conv2d: 3-7                  [6, 64, 56, 56]           [6, 256, 56, 56]          16,384\n",
       "│    │    └─BatchNorm2d: 3-8             [6, 256, 56, 56]          [6, 256, 56, 56]          512\n",
       "│    │    └─Sequential: 3-9              [6, 64, 56, 56]           [6, 256, 56, 56]          16,896\n",
       "│    │    └─ReLU: 3-10                   [6, 256, 56, 56]          [6, 256, 56, 56]          --\n",
       "│    └─Bottleneck: 2-2                   [6, 256, 56, 56]          [6, 256, 56, 56]          --\n",
       "│    │    └─Conv2d: 3-11                 [6, 256, 56, 56]          [6, 64, 56, 56]           16,384\n",
       "│    │    └─BatchNorm2d: 3-12            [6, 64, 56, 56]           [6, 64, 56, 56]           128\n",
       "│    │    └─ReLU: 3-13                   [6, 64, 56, 56]           [6, 64, 56, 56]           --\n",
       "│    │    └─Conv2d: 3-14                 [6, 64, 56, 56]           [6, 64, 56, 56]           36,864\n",
       "│    │    └─BatchNorm2d: 3-15            [6, 64, 56, 56]           [6, 64, 56, 56]           128\n",
       "│    │    └─ReLU: 3-16                   [6, 64, 56, 56]           [6, 64, 56, 56]           --\n",
       "│    │    └─Conv2d: 3-17                 [6, 64, 56, 56]           [6, 256, 56, 56]          16,384\n",
       "│    │    └─BatchNorm2d: 3-18            [6, 256, 56, 56]          [6, 256, 56, 56]          512\n",
       "│    │    └─ReLU: 3-19                   [6, 256, 56, 56]          [6, 256, 56, 56]          --\n",
       "│    └─Bottleneck: 2-3                   [6, 256, 56, 56]          [6, 256, 56, 56]          --\n",
       "│    │    └─Conv2d: 3-20                 [6, 256, 56, 56]          [6, 64, 56, 56]           16,384\n",
       "│    │    └─BatchNorm2d: 3-21            [6, 64, 56, 56]           [6, 64, 56, 56]           128\n",
       "│    │    └─ReLU: 3-22                   [6, 64, 56, 56]           [6, 64, 56, 56]           --\n",
       "│    │    └─Conv2d: 3-23                 [6, 64, 56, 56]           [6, 64, 56, 56]           36,864\n",
       "│    │    └─BatchNorm2d: 3-24            [6, 64, 56, 56]           [6, 64, 56, 56]           128\n",
       "│    │    └─ReLU: 3-25                   [6, 64, 56, 56]           [6, 64, 56, 56]           --\n",
       "│    │    └─Conv2d: 3-26                 [6, 64, 56, 56]           [6, 256, 56, 56]          16,384\n",
       "│    │    └─BatchNorm2d: 3-27            [6, 256, 56, 56]          [6, 256, 56, 56]          512\n",
       "│    │    └─ReLU: 3-28                   [6, 256, 56, 56]          [6, 256, 56, 56]          --\n",
       "├─Sequential: 1-6                        [6, 256, 56, 56]          [6, 512, 28, 28]          --\n",
       "│    └─Bottleneck: 2-4                   [6, 256, 56, 56]          [6, 512, 28, 28]          --\n",
       "│    │    └─Conv2d: 3-29                 [6, 256, 56, 56]          [6, 128, 56, 56]          32,768\n",
       "│    │    └─BatchNorm2d: 3-30            [6, 128, 56, 56]          [6, 128, 56, 56]          256\n",
       "│    │    └─ReLU: 3-31                   [6, 128, 56, 56]          [6, 128, 56, 56]          --\n",
       "│    │    └─Conv2d: 3-32                 [6, 128, 56, 56]          [6, 128, 28, 28]          147,456\n",
       "│    │    └─BatchNorm2d: 3-33            [6, 128, 28, 28]          [6, 128, 28, 28]          256\n",
       "│    │    └─ReLU: 3-34                   [6, 128, 28, 28]          [6, 128, 28, 28]          --\n",
       "│    │    └─Conv2d: 3-35                 [6, 128, 28, 28]          [6, 512, 28, 28]          65,536\n",
       "│    │    └─BatchNorm2d: 3-36            [6, 512, 28, 28]          [6, 512, 28, 28]          1,024\n",
       "│    │    └─Sequential: 3-37             [6, 256, 56, 56]          [6, 512, 28, 28]          132,096\n",
       "│    │    └─ReLU: 3-38                   [6, 512, 28, 28]          [6, 512, 28, 28]          --\n",
       "│    └─Bottleneck: 2-5                   [6, 512, 28, 28]          [6, 512, 28, 28]          --\n",
       "│    │    └─Conv2d: 3-39                 [6, 512, 28, 28]          [6, 128, 28, 28]          65,536\n",
       "│    │    └─BatchNorm2d: 3-40            [6, 128, 28, 28]          [6, 128, 28, 28]          256\n",
       "│    │    └─ReLU: 3-41                   [6, 128, 28, 28]          [6, 128, 28, 28]          --\n",
       "│    │    └─Conv2d: 3-42                 [6, 128, 28, 28]          [6, 128, 28, 28]          147,456\n",
       "│    │    └─BatchNorm2d: 3-43            [6, 128, 28, 28]          [6, 128, 28, 28]          256\n",
       "│    │    └─ReLU: 3-44                   [6, 128, 28, 28]          [6, 128, 28, 28]          --\n",
       "│    │    └─Conv2d: 3-45                 [6, 128, 28, 28]          [6, 512, 28, 28]          65,536\n",
       "│    │    └─BatchNorm2d: 3-46            [6, 512, 28, 28]          [6, 512, 28, 28]          1,024\n",
       "│    │    └─ReLU: 3-47                   [6, 512, 28, 28]          [6, 512, 28, 28]          --\n",
       "│    └─Bottleneck: 2-6                   [6, 512, 28, 28]          [6, 512, 28, 28]          --\n",
       "│    │    └─Conv2d: 3-48                 [6, 512, 28, 28]          [6, 128, 28, 28]          65,536\n",
       "│    │    └─BatchNorm2d: 3-49            [6, 128, 28, 28]          [6, 128, 28, 28]          256\n",
       "│    │    └─ReLU: 3-50                   [6, 128, 28, 28]          [6, 128, 28, 28]          --\n",
       "│    │    └─Conv2d: 3-51                 [6, 128, 28, 28]          [6, 128, 28, 28]          147,456\n",
       "│    │    └─BatchNorm2d: 3-52            [6, 128, 28, 28]          [6, 128, 28, 28]          256\n",
       "│    │    └─ReLU: 3-53                   [6, 128, 28, 28]          [6, 128, 28, 28]          --\n",
       "│    │    └─Conv2d: 3-54                 [6, 128, 28, 28]          [6, 512, 28, 28]          65,536\n",
       "│    │    └─BatchNorm2d: 3-55            [6, 512, 28, 28]          [6, 512, 28, 28]          1,024\n",
       "│    │    └─ReLU: 3-56                   [6, 512, 28, 28]          [6, 512, 28, 28]          --\n",
       "│    └─Bottleneck: 2-7                   [6, 512, 28, 28]          [6, 512, 28, 28]          --\n",
       "│    │    └─Conv2d: 3-57                 [6, 512, 28, 28]          [6, 128, 28, 28]          65,536\n",
       "│    │    └─BatchNorm2d: 3-58            [6, 128, 28, 28]          [6, 128, 28, 28]          256\n",
       "│    │    └─ReLU: 3-59                   [6, 128, 28, 28]          [6, 128, 28, 28]          --\n",
       "│    │    └─Conv2d: 3-60                 [6, 128, 28, 28]          [6, 128, 28, 28]          147,456\n",
       "│    │    └─BatchNorm2d: 3-61            [6, 128, 28, 28]          [6, 128, 28, 28]          256\n",
       "│    │    └─ReLU: 3-62                   [6, 128, 28, 28]          [6, 128, 28, 28]          --\n",
       "│    │    └─Conv2d: 3-63                 [6, 128, 28, 28]          [6, 512, 28, 28]          65,536\n",
       "│    │    └─BatchNorm2d: 3-64            [6, 512, 28, 28]          [6, 512, 28, 28]          1,024\n",
       "│    │    └─ReLU: 3-65                   [6, 512, 28, 28]          [6, 512, 28, 28]          --\n",
       "├─Sequential: 1-7                        [6, 512, 28, 28]          [6, 1024, 14, 14]         --\n",
       "│    └─Bottleneck: 2-8                   [6, 512, 28, 28]          [6, 1024, 14, 14]         --\n",
       "│    │    └─Conv2d: 3-66                 [6, 512, 28, 28]          [6, 256, 28, 28]          131,072\n",
       "│    │    └─BatchNorm2d: 3-67            [6, 256, 28, 28]          [6, 256, 28, 28]          512\n",
       "│    │    └─ReLU: 3-68                   [6, 256, 28, 28]          [6, 256, 28, 28]          --\n",
       "│    │    └─Conv2d: 3-69                 [6, 256, 28, 28]          [6, 256, 14, 14]          589,824\n",
       "│    │    └─BatchNorm2d: 3-70            [6, 256, 14, 14]          [6, 256, 14, 14]          512\n",
       "│    │    └─ReLU: 3-71                   [6, 256, 14, 14]          [6, 256, 14, 14]          --\n",
       "│    │    └─Conv2d: 3-72                 [6, 256, 14, 14]          [6, 1024, 14, 14]         262,144\n",
       "│    │    └─BatchNorm2d: 3-73            [6, 1024, 14, 14]         [6, 1024, 14, 14]         2,048\n",
       "│    │    └─Sequential: 3-74             [6, 512, 28, 28]          [6, 1024, 14, 14]         526,336\n",
       "│    │    └─ReLU: 3-75                   [6, 1024, 14, 14]         [6, 1024, 14, 14]         --\n",
       "│    └─Bottleneck: 2-9                   [6, 1024, 14, 14]         [6, 1024, 14, 14]         --\n",
       "│    │    └─Conv2d: 3-76                 [6, 1024, 14, 14]         [6, 256, 14, 14]          262,144\n",
       "│    │    └─BatchNorm2d: 3-77            [6, 256, 14, 14]          [6, 256, 14, 14]          512\n",
       "│    │    └─ReLU: 3-78                   [6, 256, 14, 14]          [6, 256, 14, 14]          --\n",
       "│    │    └─Conv2d: 3-79                 [6, 256, 14, 14]          [6, 256, 14, 14]          589,824\n",
       "│    │    └─BatchNorm2d: 3-80            [6, 256, 14, 14]          [6, 256, 14, 14]          512\n",
       "│    │    └─ReLU: 3-81                   [6, 256, 14, 14]          [6, 256, 14, 14]          --\n",
       "│    │    └─Conv2d: 3-82                 [6, 256, 14, 14]          [6, 1024, 14, 14]         262,144\n",
       "│    │    └─BatchNorm2d: 3-83            [6, 1024, 14, 14]         [6, 1024, 14, 14]         2,048\n",
       "│    │    └─ReLU: 3-84                   [6, 1024, 14, 14]         [6, 1024, 14, 14]         --\n",
       "│    └─Bottleneck: 2-10                  [6, 1024, 14, 14]         [6, 1024, 14, 14]         --\n",
       "│    │    └─Conv2d: 3-85                 [6, 1024, 14, 14]         [6, 256, 14, 14]          262,144\n",
       "│    │    └─BatchNorm2d: 3-86            [6, 256, 14, 14]          [6, 256, 14, 14]          512\n",
       "│    │    └─ReLU: 3-87                   [6, 256, 14, 14]          [6, 256, 14, 14]          --\n",
       "│    │    └─Conv2d: 3-88                 [6, 256, 14, 14]          [6, 256, 14, 14]          589,824\n",
       "│    │    └─BatchNorm2d: 3-89            [6, 256, 14, 14]          [6, 256, 14, 14]          512\n",
       "│    │    └─ReLU: 3-90                   [6, 256, 14, 14]          [6, 256, 14, 14]          --\n",
       "│    │    └─Conv2d: 3-91                 [6, 256, 14, 14]          [6, 1024, 14, 14]         262,144\n",
       "│    │    └─BatchNorm2d: 3-92            [6, 1024, 14, 14]         [6, 1024, 14, 14]         2,048\n",
       "│    │    └─ReLU: 3-93                   [6, 1024, 14, 14]         [6, 1024, 14, 14]         --\n",
       "│    └─Bottleneck: 2-11                  [6, 1024, 14, 14]         [6, 1024, 14, 14]         --\n",
       "│    │    └─Conv2d: 3-94                 [6, 1024, 14, 14]         [6, 256, 14, 14]          262,144\n",
       "│    │    └─BatchNorm2d: 3-95            [6, 256, 14, 14]          [6, 256, 14, 14]          512\n",
       "│    │    └─ReLU: 3-96                   [6, 256, 14, 14]          [6, 256, 14, 14]          --\n",
       "│    │    └─Conv2d: 3-97                 [6, 256, 14, 14]          [6, 256, 14, 14]          589,824\n",
       "│    │    └─BatchNorm2d: 3-98            [6, 256, 14, 14]          [6, 256, 14, 14]          512\n",
       "│    │    └─ReLU: 3-99                   [6, 256, 14, 14]          [6, 256, 14, 14]          --\n",
       "│    │    └─Conv2d: 3-100                [6, 256, 14, 14]          [6, 1024, 14, 14]         262,144\n",
       "│    │    └─BatchNorm2d: 3-101           [6, 1024, 14, 14]         [6, 1024, 14, 14]         2,048\n",
       "│    │    └─ReLU: 3-102                  [6, 1024, 14, 14]         [6, 1024, 14, 14]         --\n",
       "│    └─Bottleneck: 2-12                  [6, 1024, 14, 14]         [6, 1024, 14, 14]         --\n",
       "│    │    └─Conv2d: 3-103                [6, 1024, 14, 14]         [6, 256, 14, 14]          262,144\n",
       "│    │    └─BatchNorm2d: 3-104           [6, 256, 14, 14]          [6, 256, 14, 14]          512\n",
       "│    │    └─ReLU: 3-105                  [6, 256, 14, 14]          [6, 256, 14, 14]          --\n",
       "│    │    └─Conv2d: 3-106                [6, 256, 14, 14]          [6, 256, 14, 14]          589,824\n",
       "│    │    └─BatchNorm2d: 3-107           [6, 256, 14, 14]          [6, 256, 14, 14]          512\n",
       "│    │    └─ReLU: 3-108                  [6, 256, 14, 14]          [6, 256, 14, 14]          --\n",
       "│    │    └─Conv2d: 3-109                [6, 256, 14, 14]          [6, 1024, 14, 14]         262,144\n",
       "│    │    └─BatchNorm2d: 3-110           [6, 1024, 14, 14]         [6, 1024, 14, 14]         2,048\n",
       "│    │    └─ReLU: 3-111                  [6, 1024, 14, 14]         [6, 1024, 14, 14]         --\n",
       "│    └─Bottleneck: 2-13                  [6, 1024, 14, 14]         [6, 1024, 14, 14]         --\n",
       "│    │    └─Conv2d: 3-112                [6, 1024, 14, 14]         [6, 256, 14, 14]          262,144\n",
       "│    │    └─BatchNorm2d: 3-113           [6, 256, 14, 14]          [6, 256, 14, 14]          512\n",
       "│    │    └─ReLU: 3-114                  [6, 256, 14, 14]          [6, 256, 14, 14]          --\n",
       "│    │    └─Conv2d: 3-115                [6, 256, 14, 14]          [6, 256, 14, 14]          589,824\n",
       "│    │    └─BatchNorm2d: 3-116           [6, 256, 14, 14]          [6, 256, 14, 14]          512\n",
       "│    │    └─ReLU: 3-117                  [6, 256, 14, 14]          [6, 256, 14, 14]          --\n",
       "│    │    └─Conv2d: 3-118                [6, 256, 14, 14]          [6, 1024, 14, 14]         262,144\n",
       "│    │    └─BatchNorm2d: 3-119           [6, 1024, 14, 14]         [6, 1024, 14, 14]         2,048\n",
       "│    │    └─ReLU: 3-120                  [6, 1024, 14, 14]         [6, 1024, 14, 14]         --\n",
       "├─Sequential: 1-8                        [6, 1024, 14, 14]         [6, 2048, 7, 7]           --\n",
       "│    └─Bottleneck: 2-14                  [6, 1024, 14, 14]         [6, 2048, 7, 7]           --\n",
       "│    │    └─Conv2d: 3-121                [6, 1024, 14, 14]         [6, 512, 14, 14]          524,288\n",
       "│    │    └─BatchNorm2d: 3-122           [6, 512, 14, 14]          [6, 512, 14, 14]          1,024\n",
       "│    │    └─ReLU: 3-123                  [6, 512, 14, 14]          [6, 512, 14, 14]          --\n",
       "│    │    └─Conv2d: 3-124                [6, 512, 14, 14]          [6, 512, 7, 7]            2,359,296\n",
       "│    │    └─BatchNorm2d: 3-125           [6, 512, 7, 7]            [6, 512, 7, 7]            1,024\n",
       "│    │    └─ReLU: 3-126                  [6, 512, 7, 7]            [6, 512, 7, 7]            --\n",
       "│    │    └─Conv2d: 3-127                [6, 512, 7, 7]            [6, 2048, 7, 7]           1,048,576\n",
       "│    │    └─BatchNorm2d: 3-128           [6, 2048, 7, 7]           [6, 2048, 7, 7]           4,096\n",
       "│    │    └─Sequential: 3-129            [6, 1024, 14, 14]         [6, 2048, 7, 7]           2,101,248\n",
       "│    │    └─ReLU: 3-130                  [6, 2048, 7, 7]           [6, 2048, 7, 7]           --\n",
       "│    └─Bottleneck: 2-15                  [6, 2048, 7, 7]           [6, 2048, 7, 7]           --\n",
       "│    │    └─Conv2d: 3-131                [6, 2048, 7, 7]           [6, 512, 7, 7]            1,048,576\n",
       "│    │    └─BatchNorm2d: 3-132           [6, 512, 7, 7]            [6, 512, 7, 7]            1,024\n",
       "│    │    └─ReLU: 3-133                  [6, 512, 7, 7]            [6, 512, 7, 7]            --\n",
       "│    │    └─Conv2d: 3-134                [6, 512, 7, 7]            [6, 512, 7, 7]            2,359,296\n",
       "│    │    └─BatchNorm2d: 3-135           [6, 512, 7, 7]            [6, 512, 7, 7]            1,024\n",
       "│    │    └─ReLU: 3-136                  [6, 512, 7, 7]            [6, 512, 7, 7]            --\n",
       "│    │    └─Conv2d: 3-137                [6, 512, 7, 7]            [6, 2048, 7, 7]           1,048,576\n",
       "│    │    └─BatchNorm2d: 3-138           [6, 2048, 7, 7]           [6, 2048, 7, 7]           4,096\n",
       "│    │    └─ReLU: 3-139                  [6, 2048, 7, 7]           [6, 2048, 7, 7]           --\n",
       "│    └─Bottleneck: 2-16                  [6, 2048, 7, 7]           [6, 2048, 7, 7]           --\n",
       "│    │    └─Conv2d: 3-140                [6, 2048, 7, 7]           [6, 512, 7, 7]            1,048,576\n",
       "│    │    └─BatchNorm2d: 3-141           [6, 512, 7, 7]            [6, 512, 7, 7]            1,024\n",
       "│    │    └─ReLU: 3-142                  [6, 512, 7, 7]            [6, 512, 7, 7]            --\n",
       "│    │    └─Conv2d: 3-143                [6, 512, 7, 7]            [6, 512, 7, 7]            2,359,296\n",
       "│    │    └─BatchNorm2d: 3-144           [6, 512, 7, 7]            [6, 512, 7, 7]            1,024\n",
       "│    │    └─ReLU: 3-145                  [6, 512, 7, 7]            [6, 512, 7, 7]            --\n",
       "│    │    └─Conv2d: 3-146                [6, 512, 7, 7]            [6, 2048, 7, 7]           1,048,576\n",
       "│    │    └─BatchNorm2d: 3-147           [6, 2048, 7, 7]           [6, 2048, 7, 7]           4,096\n",
       "│    │    └─ReLU: 3-148                  [6, 2048, 7, 7]           [6, 2048, 7, 7]           --\n",
       "├─AdaptiveAvgPool2d: 1-9                 [6, 2048, 7, 7]           [6, 2048, 1, 1]           --\n",
       "├─Linear: 1-10                           [6, 2048]                 [6, 1000]                 2,049,000\n",
       "===================================================================================================================\n",
       "Total params: 25,557,032\n",
       "Trainable params: 25,557,032\n",
       "Non-trainable params: 0\n",
       "Total mult-adds (Units.GIGABYTES): 24.54\n",
       "===================================================================================================================\n",
       "Input size (MB): 3.61\n",
       "Forward/backward pass size (MB): 1066.99\n",
       "Params size (MB): 102.23\n",
       "Estimated Total Size (MB): 1172.83\n",
       "==================================================================================================================="
      ]
     },
     "execution_count": 63,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "resnet_model = resnet50(weights=ResNet50_Weights.DEFAULT)\n",
    "summary(resnet_model,\n",
    "        input_data=imgs,\n",
    "        col_names=['input_size',\n",
    "                   'output_size',\n",
    "                   'num_params'])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "59077750",
   "metadata": {},
   "source": [
    "We set the mode to `eval()` to ensure that the model is ready to predict on new data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "id": "9e53e23b",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ResNet(\n",
       "  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
       "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  (relu): ReLU(inplace=True)\n",
       "  (maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)\n",
       "  (layer1): Sequential(\n",
       "    (0): Bottleneck(\n",
       "      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (downsample): Sequential(\n",
       "        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      )\n",
       "    )\n",
       "    (1): Bottleneck(\n",
       "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "    )\n",
       "    (2): Bottleneck(\n",
       "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "    )\n",
       "  )\n",
       "  (layer2): Sequential(\n",
       "    (0): Bottleneck(\n",
       "      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (downsample): Sequential(\n",
       "        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
       "        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      )\n",
       "    )\n",
       "    (1): Bottleneck(\n",
       "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "    )\n",
       "    (2): Bottleneck(\n",
       "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "    )\n",
       "    (3): Bottleneck(\n",
       "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "    )\n",
       "  )\n",
       "  (layer3): Sequential(\n",
       "    (0): Bottleneck(\n",
       "      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (downsample): Sequential(\n",
       "        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
       "        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      )\n",
       "    )\n",
       "    (1): Bottleneck(\n",
       "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "    )\n",
       "    (2): Bottleneck(\n",
       "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "    )\n",
       "    (3): Bottleneck(\n",
       "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "    )\n",
       "    (4): Bottleneck(\n",
       "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "    )\n",
       "    (5): Bottleneck(\n",
       "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "    )\n",
       "  )\n",
       "  (layer4): Sequential(\n",
       "    (0): Bottleneck(\n",
       "      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "      (downsample): Sequential(\n",
       "        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
       "        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      )\n",
       "    )\n",
       "    (1): Bottleneck(\n",
       "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "    )\n",
       "    (2): Bottleneck(\n",
       "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
       "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "      (relu): ReLU(inplace=True)\n",
       "    )\n",
       "  )\n",
       "  (avgpool): AdaptiveAvgPool2d(output_size=(1, 1))\n",
       "  (fc): Linear(in_features=2048, out_features=1000, bias=True)\n",
       ")"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "resnet_model.eval()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "24d63ea9",
   "metadata": {},
   "source": [
    "Inspecting the output above, we see that when setting up the\n",
    "`resnet_model`, the authors defined a `Bottleneck`, much like our\n",
    "`BuildingBlock` module.\n",
    "\n",
    "We now feed our six images through the fitted network."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "id": "0890e08a",
   "metadata": {},
   "outputs": [],
   "source": [
    "img_preds = resnet_model(imgs)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a7d8f653",
   "metadata": {},
   "source": [
    "Let’s look at the predicted probabilities for each of the top 3 choices. First we compute\n",
    "the probabilities by applying the softmax to the logits in `img_preds`. Note that\n",
    "we have had to call the `detach()` method on the tensor `img_preds` in order to convert\n",
    "it to our a more familiar `ndarray`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "id": "60ec5152",
   "metadata": {},
   "outputs": [],
   "source": [
    "img_probs = np.exp(np.asarray(img_preds.detach()))\n",
    "img_probs /= img_probs.sum(1)[:,None]\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0ee6c41c",
   "metadata": {},
   "source": [
    "In order to see the class labels, we must download the index file associated with `imagenet`. {This is avalable from the book website and  [s3.amazonaws.com/deep-learning-models/image-models/imagenet_class_index.json](https://s3.amazonaws.com/deep-learning-models/image-models/imagenet_class_index.json).}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "id": "9408bbd3",
   "metadata": {},
   "outputs": [],
   "source": [
    "labs = json.load(open('imagenet_class_index.json'))\n",
    "class_labels = pd.DataFrame([(int(k), v[1]) for k, v in \n",
    "                           labs.items()],\n",
    "                           columns=['idx', 'label'])\n",
    "class_labels = class_labels.set_index('idx')\n",
    "class_labels = class_labels.sort_index()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4d134d92",
   "metadata": {},
   "source": [
    "We’ll now construct a data frame for each image file\n",
    "with the labels with the three highest probabilities as\n",
    "estimated by the model above."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "id": "5369faef",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Image: book_images/Cape_Weaver.jpg\n",
      "      label      prob\n",
      "0   jacamar  0.297499\n",
      "1     macaw  0.068107\n",
      "2  lorikeet  0.051105\n",
      "Image: book_images/Flamingo.jpg\n",
      "            label      prob\n",
      "0        flamingo  0.609515\n",
      "1       spoonbill  0.013586\n",
      "2  American_egret  0.002132\n",
      "Image: book_images/Hawk_Fountain.jpg\n",
      "            label      prob\n",
      "0            kite  0.184682\n",
      "1           robin  0.084021\n",
      "2  great_grey_owl  0.061274\n",
      "Image: book_images/Hawk_cropped.jpg\n",
      "            label      prob\n",
      "0            kite  0.453833\n",
      "1  great_grey_owl  0.015914\n",
      "2             jay  0.012210\n",
      "Image: book_images/Lhasa_Apso.jpg\n",
      "             label      prob\n",
      "0            Lhasa  0.260317\n",
      "1         Shih-Tzu  0.097196\n",
      "2  Tibetan_terrier  0.032820\n",
      "Image: book_images/Sleeping_Cat.jpg\n",
      "         label      prob\n",
      "0  Persian_cat  0.163070\n",
      "1        tabby  0.074143\n",
      "2    tiger_cat  0.042578\n"
     ]
    }
   ],
   "source": [
    "for i, imgfile in enumerate(imgfiles):\n",
    "    img_df = class_labels.copy()\n",
    "    img_df['prob'] = img_probs[i]\n",
    "    img_df = img_df.sort_values(by='prob', ascending=False)[:3]\n",
    "    print(f'Image: {imgfile}')\n",
    "    print(img_df.reset_index().drop(columns=['idx']))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e8da904e",
   "metadata": {},
   "source": [
    "We see that the model\n",
    "is quite confident about `Flamingo.jpg`, but a little less so for the\n",
    "other images.\n",
    "\n",
    "We end this section with our usual cleanup."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "id": "f585e5d2",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [],
   "source": [
    "del(cifar_test,\n",
    "    cifar_train,\n",
    "    cifar_dm,\n",
    "    cifar_module,\n",
    "    cifar_logger,\n",
    "    cifar_optimizer,\n",
    "    cifar_trainer)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e80c2601",
   "metadata": {},
   "source": [
    "## IMDB Document Classification\n",
    "We now implement models for sentiment classification (Section~\\ref{Ch13:sec:docum-class})  on the `IMDB`\n",
    "dataset. As mentioned above code block~8, we are using\n",
    "a preprocessed version of the `IMDB` dataset found in the\n",
    "`keras` package. As `keras` uses `tensorflow`, a different\n",
    "tensor and deep learning library, we have\n",
    "converted the data to be suitable for `torch`. The\n",
    "code used to convert from `keras` is\n",
    "available in the module `ISLP.torch._make_imdb`. It\n",
    "requires some of the `keras` packages to run. These data use a dictionary of size 10,000.\n",
    "\n",
    "We have stored three different representations of the review data for this lab:\n",
    "\n",
    "* `load_tensor()`, a sparse tensor version usable by `torch`;\n",
    "* `load_sparse()`, a sparse matrix version usable by `sklearn`, since we will compare with a lasso fit;\n",
    "* `load_sequential()`, a padded\n",
    "version of the original sequence representation, limited to the last\n",
    "500 words of each review.\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "id": "30285383",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([   1,   14,   22,   16,   43,  530,  973, 1622, 1385,   65,  458,\n",
       "       4468], dtype=int32)"
      ]
     },
     "execution_count": 70,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "(imdb_seq_train,\n",
    " imdb_seq_test) = load_sequential(root='data/IMDB')\n",
    "padded_sample = np.asarray(imdb_seq_train.tensors[0][0])\n",
    "sample_review = padded_sample[padded_sample > 0][:12]\n",
    "sample_review[:12]\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cf823309",
   "metadata": {},
   "source": [
    "The datasets `imdb_seq_train` and `imdb_seq_test` are\n",
    "both instances of the class `TensorDataset`. The\n",
    "tensors used to construct them can be found in the `tensors` attribute, with\n",
    "the first tensor  the features `X` and the second  the outcome `Y`.\n",
    "We have taken the first row of features and stored it as `padded_sample`. In the preprocessing\n",
    "used to form these data, sequences were padded with 0s in the beginning if they were\n",
    "not long enough, hence we remove this padding by restricting to entries where\n",
    "`padded_sample > 0`. We then provide the first 12 words of the sample review.\n",
    "\n",
    "We can find these words in the `lookup` dictionary from the `ISLP.torch.imdb` module."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "id": "0bfe31c0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "\"<START> this film was just brilliant casting location scenery story direction everyone's\""
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lookup = load_lookup(root='data/IMDB')\n",
    "' '.join(lookup[i] for i in sample_review)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "be8794d4",
   "metadata": {},
   "source": [
    "For our first model, we have created a binary feature for each\n",
    "of the 10,000 possible words in the dataset, with an entry of one\n",
    "in the $i,j$ entry if word $j$ appears in review $i$. As most reviews\n",
    "are quite short, such a feature matrix has over 98% zeros. These data\n",
    "are accessed using `load_tensor()` from the `ISLP` library."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "id": "46a4749f",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "max_num_workers=10\n",
    "(imdb_train,\n",
    " imdb_test) = load_tensor(root='data/IMDB')\n",
    "imdb_dm = SimpleDataModule(imdb_train,\n",
    "                           imdb_test,\n",
    "                           validation=2000,\n",
    "                           num_workers=min(6, max_num_workers),\n",
    "                           batch_size=512)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b1772177",
   "metadata": {},
   "source": [
    "We’ll use a two-layer model for our first model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "id": "7400ab60",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "class IMDBModel(nn.Module):\n",
    "\n",
    "    def __init__(self, input_size):\n",
    "        super(IMDBModel, self).__init__()\n",
    "        self.dense1 = nn.Linear(input_size, 16)\n",
    "        self.activation = nn.ReLU()\n",
    "        self.dense2 = nn.Linear(16, 16)\n",
    "        self.output = nn.Linear(16, 1)\n",
    "\n",
    "    def forward(self, x):\n",
    "        val = x\n",
    "        for _map in [self.dense1,\n",
    "                     self.activation,\n",
    "                     self.dense2,\n",
    "                     self.activation,\n",
    "                     self.output]:\n",
    "            val = _map(val)\n",
    "        return torch.flatten(val)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f4596108",
   "metadata": {},
   "source": [
    "We now instantiate our model and look at a summary."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "id": "51960f4e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "===================================================================================================================\n",
       "Layer (type:depth-idx)                   Input Shape               Output Shape              Param #\n",
       "===================================================================================================================\n",
       "IMDBModel                                [25000, 10003]            [25000]                   --\n",
       "├─Linear: 1-1                            [25000, 10003]            [25000, 16]               160,064\n",
       "├─ReLU: 1-2                              [25000, 16]               [25000, 16]               --\n",
       "├─Linear: 1-3                            [25000, 16]               [25000, 16]               272\n",
       "├─ReLU: 1-4                              [25000, 16]               [25000, 16]               --\n",
       "├─Linear: 1-5                            [25000, 16]               [25000, 1]                17\n",
       "===================================================================================================================\n",
       "Total params: 160,353\n",
       "Trainable params: 160,353\n",
       "Non-trainable params: 0\n",
       "Total mult-adds (Units.GIGABYTES): 4.01\n",
       "===================================================================================================================\n",
       "Input size (MB): 1000.30\n",
       "Forward/backward pass size (MB): 6.60\n",
       "Params size (MB): 0.64\n",
       "Estimated Total Size (MB): 1007.54\n",
       "==================================================================================================================="
      ]
     },
     "execution_count": 74,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "imdb_model = IMDBModel(imdb_test.tensors[0].size()[1])\n",
    "summary(imdb_model,\n",
    "        input_size=imdb_test.tensors[0].size(),\n",
    "        col_names=['input_size',\n",
    "                   'output_size',\n",
    "                   'num_params'])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6433c5a4",
   "metadata": {},
   "source": [
    "We’ll again use\n",
    "a smaller learning rate for these data,\n",
    "hence we pass an `optimizer` to the\n",
    "`SimpleModule`. \n",
    "Since the reviews are classified into\n",
    "positive or negative sentiment, we use\n",
    "`SimpleModule.binary_classification()`. {Our use of\n",
    "  `binary_classification()` instead of  `classification()` is\n",
    "  due to some subtlety in how `torchmetrics.Accuracy()` works,\n",
    "as well as the data type of the targets.}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "id": "75faa2be",
   "metadata": {},
   "outputs": [],
   "source": [
    "imdb_optimizer = RMSprop(imdb_model.parameters(), lr=0.001)\n",
    "imdb_module = SimpleModule.binary_classification(\n",
    "                         imdb_model,\n",
    "                         optimizer=imdb_optimizer)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "71fbb840",
   "metadata": {},
   "source": [
    "Having loaded the datasets into a data module\n",
    "and created a `SimpleModule`, the remaining steps\n",
    "are familiar."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "id": "2637adf5",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "GPU available: True (mps), used: True\n",
      "TPU available: False, using: 0 TPU cores\n",
      "IPU available: False, using: 0 IPUs\n",
      "HPU available: False, using: 0 HPUs\n",
      "\n",
      "  | Name  | Type              | Params\n",
      "--------------------------------------------\n",
      "0 | model | IMDBModel         | 160 K \n",
      "1 | loss  | BCEWithLogitsLoss | 0     \n",
      "--------------------------------------------\n",
      "160 K     Trainable params\n",
      "0         Non-trainable params\n",
      "160 K     Total params\n",
      "0.641     Total estimated model params size (MB)\n",
      "/Users/jtaylo/anaconda3/envs/ISLP_v22_312/lib/python3.12/site-packages/pytorch_lightning/loops/fit_loop.py:298: The number of training batches (45) is smaller than the logging interval Trainer(log_every_n_steps=50). Set a lower value for log_every_n_steps if you want to see logs for the training epoch.\n",
      "`Trainer.fit` stopped: `max_epochs=30` reached.\n"
     ]
    }
   ],
   "source": [
    "imdb_logger = CSVLogger('logs', name='IMDB')\n",
    "imdb_trainer = Trainer(deterministic=True,\n",
    "                       max_epochs=30,\n",
    "                       logger=imdb_logger,\n",
    "                       enable_progress_bar=False,\n",
    "                       callbacks=[ErrorTracker()])\n",
    "imdb_trainer.fit(imdb_module,\n",
    "                 datamodule=imdb_dm)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e077efc8",
   "metadata": {},
   "source": [
    "Evaluating the test error yields roughly 86% accuracy."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "id": "0500cc62",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\">        Test metric        </span>┃<span style=\"font-weight: bold\">       DataLoader 0        </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">       test_accuracy       </span>│<span style=\"color: #800080; text-decoration-color: #800080\">    0.8450000286102295     </span>│\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">         test_loss         </span>│<span style=\"color: #800080; text-decoration-color: #800080\">    1.2167928218841553     </span>│\n",
       "└───────────────────────────┴───────────────────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1m       Test metric       \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      DataLoader 0       \u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│\u001b[36m \u001b[0m\u001b[36m      test_accuracy      \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m   0.8450000286102295    \u001b[0m\u001b[35m \u001b[0m│\n",
       "│\u001b[36m \u001b[0m\u001b[36m        test_loss        \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m   1.2167928218841553    \u001b[0m\u001b[35m \u001b[0m│\n",
       "└───────────────────────────┴───────────────────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "[{'test_loss': 1.2167928218841553, 'test_accuracy': 0.8450000286102295}]"
      ]
     },
     "execution_count": 77,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_results = imdb_trainer.test(imdb_module, datamodule=imdb_dm)\n",
    "test_results"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f73c7526",
   "metadata": {},
   "source": [
    "### Comparison to Lasso\n",
    "We now fit a lasso logistic regression model\n",
    " using `LogisticRegression()` from `sklearn`. Since `sklearn` does not recognize\n",
    "the sparse tensors of `torch`, we use a sparse\n",
    "matrix that is recognized by `sklearn.`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "id": "a6e81d74",
   "metadata": {},
   "outputs": [],
   "source": [
    "((X_train, Y_train),\n",
    " (X_valid, Y_valid),\n",
    " (X_test, Y_test)) = load_sparse(validation=2000,\n",
    "                                 random_state=0,\n",
    "                                 root='data/IMDB')\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3be8b857",
   "metadata": {},
   "source": [
    "Similar to what we did in\n",
    "Section~\\ref{Ch13-deeplearning-lab:single-layer-network-on-hitters-data},\n",
    "we construct a series of 50 values for the lasso reguralization parameter $\\lambda$."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "id": "b3049398",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "lam_max = np.abs(X_train.T * (Y_train - Y_train.mean())).max()\n",
    "lam_val = lam_max * np.exp(np.linspace(np.log(1),\n",
    "                                       np.log(1e-4), 50))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f08819f7",
   "metadata": {},
   "source": [
    "With `LogisticRegression()` the regularization parameter\n",
    "$C$ is specified as the inverse of $\\lambda$. There are several\n",
    "solvers for logistic regression; here we use `liblinear` which\n",
    "works well with the sparse input format. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "id": "e075f0b6",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "logit = LogisticRegression(penalty='l1', \n",
    "                           C=1/lam_max,\n",
    "                           solver='liblinear',\n",
    "                           warm_start=True,\n",
    "                           fit_intercept=True)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "94c8b4ff",
   "metadata": {},
   "source": [
    "The path of 50 values takes approximately 40 seconds to run."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "id": "a546c39a",
   "metadata": {},
   "outputs": [],
   "source": [
    "coefs = []\n",
    "intercepts = []\n",
    "\n",
    "for l in lam_val:\n",
    "    logit.C = 1/l\n",
    "    logit.fit(X_train, Y_train)\n",
    "    coefs.append(logit.coef_.copy())\n",
    "    intercepts.append(logit.intercept_)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5a33b90e",
   "metadata": {},
   "source": [
    "The coefficient and intercepts have an extraneous dimension which can be removed\n",
    "by the `np.squeeze()`  function."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "id": "2190268a",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "coefs = np.squeeze(coefs)\n",
    "intercepts = np.squeeze(intercepts)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1271d2a2",
   "metadata": {},
   "source": [
    "We’ll now make a plot to compare our neural network results with the\n",
    "lasso."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "id": "39221738",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "%%capture\n",
    "fig, axes = subplots(1, 2, figsize=(16, 8), sharey=True)\n",
    "for ((X_, Y_),\n",
    "     data_,\n",
    "     color) in zip([(X_train, Y_train),\n",
    "                    (X_valid, Y_valid),\n",
    "                    (X_test, Y_test)],\n",
    "                    ['Training', 'Validation', 'Test'],\n",
    "                    ['black', 'red', 'blue']):\n",
    "    linpred_ = X_ * coefs.T + intercepts[None,:]\n",
    "    label_ = np.array(linpred_ > 0)\n",
    "    accuracy_ = np.array([np.mean(Y_ == l) for l in label_.T])\n",
    "    axes[0].plot(-np.log(lam_val / X_train.shape[0]),\n",
    "                 accuracy_,\n",
    "                 '.--',\n",
    "                 color=color,\n",
    "                 markersize=13,\n",
    "                 linewidth=2,\n",
    "                 label=data_)\n",
    "axes[0].legend()\n",
    "axes[0].set_xlabel(r'$-\\log(\\lambda)$', fontsize=20)\n",
    "axes[0].set_ylabel('Accuracy', fontsize=20)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d40e3d0e",
   "metadata": {},
   "source": [
    "Notice the use of `%%capture`, which suppresses the displaying of the partially completed figure. This is useful\n",
    "when making a complex figure, since the steps can be spread across two or more cells.\n",
    "We now add a plot of the lasso accuracy, and display the composed figure by simply entering its name at the end of the cell."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "id": "9ba471b4",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1600x800 with 2 Axes>"
      ]
     },
     "execution_count": 84,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "imdb_results = pd.read_csv(imdb_logger.experiment.metrics_file_path)\n",
    "summary_plot(imdb_results,\n",
    "             axes[1],\n",
    "             col='accuracy',\n",
    "             ylabel='Accuracy')\n",
    "axes[1].set_xticks(np.linspace(0, 30, 7).astype(int))\n",
    "axes[1].set_ylabel('Accuracy', fontsize=20)\n",
    "axes[1].set_xlabel('Epoch', fontsize=20)\n",
    "axes[1].set_ylim([0.5, 1]);\n",
    "axes[1].axhline(test_results[0]['test_accuracy'],\n",
    "                color='blue',\n",
    "                linestyle='--',\n",
    "                linewidth=3)\n",
    "fig"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "95e2c89f",
   "metadata": {},
   "source": [
    "From the graphs we see that the accuracy of the lasso logistic regression peaks at about $0.88$,  as it does for  the neural network.\n",
    "\n",
    "Once again, we end with a cleanup."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "id": "5f8c7df0",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [],
   "source": [
    "del(imdb_model,\n",
    "    imdb_trainer,\n",
    "    imdb_logger,\n",
    "    imdb_dm,\n",
    "    imdb_train,\n",
    "    imdb_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d2a9982b",
   "metadata": {},
   "source": [
    "## Recurrent Neural Networks\n",
    "In this lab we fit the models illustrated in\n",
    "Section~\\ref{Ch13:sec:recurr-neur-netw}."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "51642336",
   "metadata": {},
   "source": [
    "### Sequential Models for Document Classification\n",
    "Here we  fit a simple  LSTM RNN for sentiment prediction to\n",
    "the `IMDb` movie-review data, as discussed in Section~\\ref{Ch13:sec:sequ-models-docum}.\n",
    "For an RNN we use the sequence of words in a document, taking their\n",
    "order into account. We loaded the preprocessed \n",
    "data at the beginning of\n",
    "Section~\\ref{Ch13-deeplearning-lab:imdb-document-classification}.\n",
    "A script that details the preprocessing can be found in the\n",
    "`ISLP` library. Notably, since more than 90% of the documents\n",
    "had fewer than 500 words, we set the document length to 500. For\n",
    "longer documents, we used the last 500 words, and for shorter\n",
    "documents, we padded the front with blanks.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "id": "af4d3414",
   "metadata": {},
   "outputs": [],
   "source": [
    "imdb_seq_dm = SimpleDataModule(imdb_seq_train,\n",
    "                               imdb_seq_test,\n",
    "                               validation=2000,\n",
    "                               batch_size=300,\n",
    "                               num_workers=min(6, max_num_workers)\n",
    "                               )\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3aa82266",
   "metadata": {},
   "source": [
    "The first layer of the RNN is an embedding layer of size 32, which will be\n",
    "learned during  training. This layer one-hot encodes  each document\n",
    "as a matrix of dimension $500 \\times 10,003$, and then maps these\n",
    "$10,003$ dimensions down to $32$. {The extra 3 dimensions\n",
    "correspond to commonly occurring non-word entries in the reviews.}\n",
    " Since each word is represented by an\n",
    "integer, this is effectively achieved by the creation of an embedding\n",
    "matrix of size $10,003\\times 32$; each of the 500 integers in the\n",
    "document are then mapped to the appropriate 32 real numbers by\n",
    "indexing the appropriate rows of this matrix."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ed2d5d9b",
   "metadata": {},
   "source": [
    "The second  layer is an LSTM with 32 units, and the output\n",
    "layer is a single logit for the binary classification task.\n",
    "In the last line of the `forward()` method below,\n",
    "we take the last 32-dimensional output of the LSTM and map it to our response."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "id": "053cda21",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "class LSTMModel(nn.Module):\n",
    "    def __init__(self, input_size):\n",
    "        super(LSTMModel, self).__init__()\n",
    "        self.embedding = nn.Embedding(input_size, 32)\n",
    "        self.lstm = nn.LSTM(input_size=32,\n",
    "                            hidden_size=32,\n",
    "                            batch_first=True)\n",
    "        self.dense = nn.Linear(32, 1)\n",
    "    def forward(self, x):\n",
    "        val, (h_n, c_n) = self.lstm(self.embedding(x))\n",
    "        return torch.flatten(self.dense(val[:,-1]))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "aad0a983",
   "metadata": {},
   "source": [
    "We instantiate and take a look at the summary of the model, using the\n",
    "first 10 documents in the corpus."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "id": "1375cc01",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "===================================================================================================================\n",
       "Layer (type:depth-idx)                   Input Shape               Output Shape              Param #\n",
       "===================================================================================================================\n",
       "LSTMModel                                [10, 500]                 [10]                      --\n",
       "├─Embedding: 1-1                         [10, 500]                 [10, 500, 32]             320,096\n",
       "├─LSTM: 1-2                              [10, 500, 32]             [10, 500, 32]             8,448\n",
       "├─Linear: 1-3                            [10, 32]                  [10, 1]                   33\n",
       "===================================================================================================================\n",
       "Total params: 328,577\n",
       "Trainable params: 328,577\n",
       "Non-trainable params: 0\n",
       "Total mult-adds (Units.MEGABYTES): 45.44\n",
       "===================================================================================================================\n",
       "Input size (MB): 50.00\n",
       "Forward/backward pass size (MB): 2.56\n",
       "Params size (MB): 1.31\n",
       "Estimated Total Size (MB): 53.87\n",
       "==================================================================================================================="
      ]
     },
     "execution_count": 88,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lstm_model = LSTMModel(X_test.shape[-1])\n",
    "summary(lstm_model,\n",
    "        input_data=imdb_seq_train.tensors[0][:10],\n",
    "        col_names=['input_size',\n",
    "                   'output_size',\n",
    "                   'num_params'])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fa9049a4",
   "metadata": {},
   "source": [
    "The 10,003 is suppressed in the summary, but we see it in the\n",
    "parameter count, since $10,003\\times 32=320,096$."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "id": "f89f58af",
   "metadata": {},
   "outputs": [],
   "source": [
    "lstm_module = SimpleModule.binary_classification(lstm_model)\n",
    "lstm_logger = CSVLogger('logs', name='IMDB_LSTM')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "id": "ae8fc1fe",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "GPU available: True (mps), used: True\n",
      "TPU available: False, using: 0 TPU cores\n",
      "IPU available: False, using: 0 IPUs\n",
      "HPU available: False, using: 0 HPUs\n",
      "\n",
      "  | Name  | Type              | Params\n",
      "--------------------------------------------\n",
      "0 | model | LSTMModel         | 328 K \n",
      "1 | loss  | BCEWithLogitsLoss | 0     \n",
      "--------------------------------------------\n",
      "328 K     Trainable params\n",
      "0         Non-trainable params\n",
      "328 K     Total params\n",
      "1.314     Total estimated model params size (MB)\n",
      "`Trainer.fit` stopped: `max_epochs=20` reached.\n"
     ]
    }
   ],
   "source": [
    "lstm_trainer = Trainer(deterministic=True,\n",
    "                       max_epochs=20,\n",
    "                       logger=lstm_logger,\n",
    "                       enable_progress_bar=False,\n",
    "                       callbacks=[ErrorTracker()])\n",
    "lstm_trainer.fit(lstm_module,\n",
    "                 datamodule=imdb_seq_dm)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cf6061b3",
   "metadata": {},
   "source": [
    "The rest is now similar to other networks we have fit. We\n",
    "track the test performance as the network is fit, and see that it attains 85% accuracy."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "id": "9c8dd061",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\">        Test metric        </span>┃<span style=\"font-weight: bold\">       DataLoader 0        </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">       test_accuracy       </span>│<span style=\"color: #800080; text-decoration-color: #800080\">    0.8505200147628784     </span>│\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">         test_loss         </span>│<span style=\"color: #800080; text-decoration-color: #800080\">    0.7480539679527283     </span>│\n",
       "└───────────────────────────┴───────────────────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1m       Test metric       \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      DataLoader 0       \u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│\u001b[36m \u001b[0m\u001b[36m      test_accuracy      \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m   0.8505200147628784    \u001b[0m\u001b[35m \u001b[0m│\n",
       "│\u001b[36m \u001b[0m\u001b[36m        test_loss        \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m   0.7480539679527283    \u001b[0m\u001b[35m \u001b[0m│\n",
       "└───────────────────────────┴───────────────────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "[{'test_loss': 0.7480539679527283, 'test_accuracy': 0.8505200147628784}]"
      ]
     },
     "execution_count": 91,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lstm_trainer.test(lstm_module, datamodule=imdb_seq_dm)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "40823178",
   "metadata": {},
   "source": [
    "We once again show the learning progress, followed by cleanup."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "id": "774197a2",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.5, 1.0)"
      ]
     },
     "execution_count": 92,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 600x600 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "lstm_results = pd.read_csv(lstm_logger.experiment.metrics_file_path)\n",
    "fig, ax = subplots(1, 1, figsize=(6, 6))\n",
    "summary_plot(lstm_results,\n",
    "             ax,\n",
    "             col='accuracy',\n",
    "             ylabel='Accuracy')\n",
    "ax.set_xticks(np.linspace(0, 20, 5).astype(int))\n",
    "ax.set_ylabel('Accuracy')\n",
    "ax.set_ylim([0.5, 1])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "id": "dbae5960",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [],
   "source": [
    "del(lstm_model,\n",
    "    lstm_trainer,\n",
    "    lstm_logger,\n",
    "    imdb_seq_dm,\n",
    "    imdb_seq_train,\n",
    "    imdb_seq_test)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5d8a3c8f",
   "metadata": {},
   "source": [
    "### Time Series Prediction\n",
    "We now show how to fit the models in Section~\\ref{Ch13:sec:time-seri-pred}\n",
    "for  time series prediction.\n",
    "We first load and standardize the data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "id": "023c7d14",
   "metadata": {},
   "outputs": [],
   "source": [
    "NYSE = load_data('NYSE')\n",
    "cols = ['DJ_return', 'log_volume', 'log_volatility']\n",
    "X = pd.DataFrame(StandardScaler(\n",
    "                     with_mean=True,\n",
    "                     with_std=True).fit_transform(NYSE[cols]),\n",
    "                 columns=NYSE[cols].columns,\n",
    "                 index=NYSE.index)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "03d5171b",
   "metadata": {},
   "source": [
    "Next we set up the lagged versions of the data, dropping\n",
    "any rows with missing values using the `dropna()`  method."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "id": "24aa4a0a",
   "metadata": {},
   "outputs": [],
   "source": [
    "for lag in range(1, 6):\n",
    "    for col in cols:\n",
    "        newcol = np.zeros(X.shape[0]) * np.nan\n",
    "        newcol[lag:] = X[col].values[:-lag]\n",
    "        X.insert(len(X.columns), \"{0}_{1}\".format(col, lag), newcol)\n",
    "X.insert(len(X.columns), 'train', NYSE['train'])\n",
    "X = X.dropna()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e2a02e25",
   "metadata": {},
   "source": [
    "Finally, we extract the response, training indicator, and drop the current day’s `DJ_return` and\n",
    "`log_volatility` to predict only from previous day’s data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "id": "5f0a4c0e",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['DJ_return_1', 'log_volume_1', 'log_volatility_1', 'DJ_return_2',\n",
       "       'log_volume_2', 'log_volatility_2', 'DJ_return_3', 'log_volume_3',\n",
       "       'log_volatility_3', 'DJ_return_4', 'log_volume_4', 'log_volatility_4',\n",
       "       'DJ_return_5', 'log_volume_5', 'log_volatility_5'],\n",
       "      dtype='object')"
      ]
     },
     "execution_count": 96,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Y, train = X['log_volume'], X['train']\n",
    "X = X.drop(columns=['train'] + cols)\n",
    "X.columns\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6bc66b8e",
   "metadata": {},
   "source": [
    "We first fit a simple linear model and compute the $R^2$ on the test data using\n",
    "the `score()`  method."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "id": "4ec80811",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.4128912938562521"
      ]
     },
     "execution_count": 97,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "M = LinearRegression()\n",
    "M.fit(X[train], Y[train])\n",
    "M.score(X[~train], Y[~train])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3053ae12",
   "metadata": {},
   "source": [
    "We refit this model, including the factor variable `day_of_week`.\n",
    "For a categorical series in `pandas`, we can form the indicators\n",
    "using the `get_dummies()`  method."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "id": "ec2de6ca",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "X_day = pd.concat([X, \n",
    "                  pd.get_dummies(NYSE['day_of_week'])],\n",
    "                  axis=1).dropna()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "765b70b4",
   "metadata": {},
   "source": [
    " Note that we do not have\n",
    "to reinstantiate the linear regression model\n",
    "as its `fit()` method accepts a design matrix and a response directly."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "id": "8ff00b65",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.4595563133053273"
      ]
     },
     "execution_count": 99,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "M.fit(X_day[train], Y[train])\n",
    "M.score(X_day[~train], Y[~train])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "88374112",
   "metadata": {},
   "source": [
    "This model achieves an $R^2$ of about 46%."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f2160545",
   "metadata": {},
   "source": [
    "To fit the RNN, we must reshape the data, as it will expect 5 lagged\n",
    "versions of each feature as indicated by the  `input_shape` argument\n",
    "to the layer `nn.RNN()`  below. We first\n",
    "ensure the columns of  our data frame are such that a reshaped\n",
    "matrix will have the variables correctly lagged. We use the\n",
    "`reindex()`  method to do this.\n",
    "\n",
    "For an input shape `(5,3)`, each row represents a lagged version of the three variables.\n",
    "The `nn.RNN()` layer also expects the first row of each\n",
    "observation to be earliest in time, so we must reverse the current order.  Hence we loop over\n",
    "`range(5,0,-1)` below, which is\n",
    "an example of using a  `slice()`  to index\n",
    "iterable objects. The general notation is `start:end:step`. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "id": "08107439",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['DJ_return_5', 'log_volume_5', 'log_volatility_5', 'DJ_return_4',\n",
       "       'log_volume_4', 'log_volatility_4', 'DJ_return_3', 'log_volume_3',\n",
       "       'log_volatility_3', 'DJ_return_2', 'log_volume_2', 'log_volatility_2',\n",
       "       'DJ_return_1', 'log_volume_1', 'log_volatility_1'],\n",
       "      dtype='object')"
      ]
     },
     "execution_count": 100,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ordered_cols = []\n",
    "for lag in range(5,0,-1):\n",
    "    for col in cols:\n",
    "        ordered_cols.append('{0}_{1}'.format(col, lag))\n",
    "X = X.reindex(columns=ordered_cols)\n",
    "X.columns\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d7f38634",
   "metadata": {},
   "source": [
    "We now reshape the data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "id": "1776f3d2",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(6046, 5, 3)"
      ]
     },
     "execution_count": 101,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_rnn = X.to_numpy().reshape((-1,5,3))\n",
    "X_rnn.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "385f3387",
   "metadata": {},
   "source": [
    "By specifying the first size as -1, `numpy.reshape()` deduces its size based on the remaining arguments.\n",
    "\n",
    "Now we are ready to proceed with the RNN, which uses 12 hidden units, and  10%\n",
    "dropout.\n",
    "After passing through the RNN,  we extract the final time point as `val[:,-1]`\n",
    "in `forward()` below. This gets passed through a 10% dropout and then flattened through\n",
    "a linear layer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "id": "355e86eb",
   "metadata": {},
   "outputs": [],
   "source": [
    "class NYSEModel(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(NYSEModel, self).__init__()\n",
    "        self.rnn = nn.RNN(3,\n",
    "                          12,\n",
    "                          batch_first=True)\n",
    "        self.dense = nn.Linear(12, 1)\n",
    "        self.dropout = nn.Dropout(0.1)\n",
    "    def forward(self, x):\n",
    "        val, h_n = self.rnn(x)\n",
    "        val = self.dense(self.dropout(val[:,-1]))\n",
    "        return torch.flatten(val)\n",
    "nyse_model = NYSEModel()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e3abd764",
   "metadata": {},
   "source": [
    "We  fit the model in a similar fashion to previous networks. We\n",
    "supply the `fit` function with test data as validation data, so that when\n",
    "we monitor its progress and plot the history function we can see the\n",
    "progress on the test data. Of course we should not use this as a basis for\n",
    "early stopping, since then the test performance would be biased.\n",
    "\n",
    "We form the training dataset similar to\n",
    "our `Hitters` example."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "id": "72ab9a66",
   "metadata": {},
   "outputs": [],
   "source": [
    "datasets = []\n",
    "for mask in [train, ~train]:\n",
    "    X_rnn_t = torch.tensor(X_rnn[mask].astype(np.float32))\n",
    "    Y_t = torch.tensor(Y[mask].astype(np.float32))\n",
    "    datasets.append(TensorDataset(X_rnn_t, Y_t))\n",
    "nyse_train, nyse_test = datasets\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b81c1f1e",
   "metadata": {},
   "source": [
    "Following our usual pattern, we inspect the summary."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "id": "1c10263b",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "===================================================================================================================\n",
       "Layer (type:depth-idx)                   Input Shape               Output Shape              Param #\n",
       "===================================================================================================================\n",
       "NYSEModel                                [1770, 5, 3]              [1770]                    --\n",
       "├─RNN: 1-1                               [1770, 5, 3]              [1770, 5, 12]             204\n",
       "├─Dropout: 1-2                           [1770, 12]                [1770, 12]                --\n",
       "├─Linear: 1-3                            [1770, 12]                [1770, 1]                 13\n",
       "===================================================================================================================\n",
       "Total params: 217\n",
       "Trainable params: 217\n",
       "Non-trainable params: 0\n",
       "Total mult-adds (Units.MEGABYTES): 1.83\n",
       "===================================================================================================================\n",
       "Input size (MB): 0.11\n",
       "Forward/backward pass size (MB): 0.86\n",
       "Params size (MB): 0.00\n",
       "Estimated Total Size (MB): 0.97\n",
       "==================================================================================================================="
      ]
     },
     "execution_count": 104,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "summary(nyse_model,\n",
    "        input_data=X_rnn_t,\n",
    "        col_names=['input_size',\n",
    "                   'output_size',\n",
    "                   'num_params'])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0ba24d5c",
   "metadata": {},
   "source": [
    "We again put the two datasets into a data module, with a\n",
    "batch size of 64."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "id": "c36032ef",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [],
   "source": [
    "nyse_dm = SimpleDataModule(nyse_train,\n",
    "                           nyse_test,\n",
    "                           num_workers=min(4, max_num_workers),\n",
    "                           validation=nyse_test,\n",
    "                           batch_size=64)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "719a7cf8",
   "metadata": {},
   "source": [
    "We run some data through our model to be sure the sizes match up correctly."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "id": "90df0c75",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([64]) torch.Size([64])\n",
      "torch.Size([64]) torch.Size([64])\n",
      "torch.Size([64]) torch.Size([64])\n"
     ]
    }
   ],
   "source": [
    "for idx, (x, y) in enumerate(nyse_dm.train_dataloader()):\n",
    "    out = nyse_model(x)\n",
    "    print(y.size(), out.size())\n",
    "    if idx >= 2:\n",
    "        break\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a85106c8",
   "metadata": {},
   "source": [
    "We follow our previous example for setting up a trainer for a\n",
    "regression problem, requesting the $R^2$ metric\n",
    "to be be computed at each epoch."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "id": "064fcd0d",
   "metadata": {},
   "outputs": [],
   "source": [
    "nyse_optimizer = RMSprop(nyse_model.parameters(),\n",
    "                         lr=0.001)\n",
    "nyse_module = SimpleModule.regression(nyse_model,\n",
    "                                      optimizer=nyse_optimizer,\n",
    "                                      metrics={'r2':R2Score()})\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cf9a430a",
   "metadata": {},
   "source": [
    "Fitting the model should by now be familiar.\n",
    "The results on the test data are very similar to the linear AR model. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 108,
   "id": "95ff00af",
   "metadata": {
    "lines_to_next_cell": 2
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "GPU available: True (mps), used: True\n",
      "TPU available: False, using: 0 TPU cores\n",
      "IPU available: False, using: 0 IPUs\n",
      "HPU available: False, using: 0 HPUs\n",
      "\n",
      "  | Name  | Type      | Params\n",
      "------------------------------------\n",
      "0 | model | NYSEModel | 217   \n",
      "1 | loss  | MSELoss   | 0     \n",
      "------------------------------------\n",
      "217       Trainable params\n",
      "0         Non-trainable params\n",
      "217       Total params\n",
      "0.001     Total estimated model params size (MB)\n",
      "`Trainer.fit` stopped: `max_epochs=200` reached.\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\">        Test metric        </span>┃<span style=\"font-weight: bold\">       DataLoader 0        </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">         test_loss         </span>│<span style=\"color: #800080; text-decoration-color: #800080\">     0.616382360458374     </span>│\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">          test_r2          </span>│<span style=\"color: #800080; text-decoration-color: #800080\">    0.4150230884552002     </span>│\n",
       "└───────────────────────────┴───────────────────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1m       Test metric       \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      DataLoader 0       \u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│\u001b[36m \u001b[0m\u001b[36m        test_loss        \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m    0.616382360458374    \u001b[0m\u001b[35m \u001b[0m│\n",
       "│\u001b[36m \u001b[0m\u001b[36m         test_r2         \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m   0.4150230884552002    \u001b[0m\u001b[35m \u001b[0m│\n",
       "└───────────────────────────┴───────────────────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "[{'test_loss': 0.616382360458374, 'test_r2': 0.4150230884552002}]"
      ]
     },
     "execution_count": 108,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "nyse_trainer = Trainer(deterministic=True,\n",
    "                       max_epochs=200,\n",
    "                       enable_progress_bar=False,\n",
    "                       callbacks=[ErrorTracker()])\n",
    "nyse_trainer.fit(nyse_module,\n",
    "                 datamodule=nyse_dm)\n",
    "nyse_trainer.test(nyse_module,\n",
    "                  datamodule=nyse_dm)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8a0fb76e",
   "metadata": {},
   "source": [
    "We could also fit a model without the `nn.RNN()` layer by just\n",
    "using a `nn.Flatten()` layer instead. This would be a nonlinear AR model. If in addition we excluded the \n",
    "hidden layer, this would be equivalent to our earlier linear AR model.  \n",
    "\n",
    "Instead we will fit a nonlinear AR model using the feature set `X_day` that includes the `day_of_week` indicators.\n",
    "To do so, we\n",
    "must first create our test and training datasets and a corresponding\n",
    "data module. This may seem a little burdensome, but is part of the\n",
    "general pipeline for `torch`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "id": "6b1b56fc",
   "metadata": {},
   "outputs": [],
   "source": [
    "datasets = []\n",
    "for mask in [train, ~train]:\n",
    "    X_day_t = torch.tensor(\n",
    "                   np.asarray(X_day[mask]).astype(np.float32))\n",
    "    Y_t = torch.tensor(np.asarray(Y[mask]).astype(np.float32))\n",
    "    datasets.append(TensorDataset(X_day_t, Y_t))\n",
    "day_train, day_test = datasets"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "75ea59a7",
   "metadata": {},
   "source": [
    "Creating a data module follows a familiar pattern."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 110,
   "id": "10c70d0a",
   "metadata": {},
   "outputs": [],
   "source": [
    "day_dm = SimpleDataModule(day_train,\n",
    "                          day_test,\n",
    "                          num_workers=min(4, max_num_workers),\n",
    "                          validation=day_test,\n",
    "                          batch_size=64)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9c2a76bb",
   "metadata": {},
   "source": [
    "We build a `NonLinearARModel()` that takes as input the 20 features and a hidden layer  with 32 units. The remaining steps are familiar."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "id": "42fca594",
   "metadata": {},
   "outputs": [],
   "source": [
    "class NonLinearARModel(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(NonLinearARModel, self).__init__()\n",
    "        self._forward = nn.Sequential(nn.Flatten(),\n",
    "                                      nn.Linear(20, 32),\n",
    "                                      nn.ReLU(),\n",
    "                                      nn.Dropout(0.5),\n",
    "                                      nn.Linear(32, 1))\n",
    "    def forward(self, x):\n",
    "        return torch.flatten(self._forward(x))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "id": "9a134064",
   "metadata": {},
   "outputs": [],
   "source": [
    "nl_model = NonLinearARModel()\n",
    "nl_optimizer = RMSprop(nl_model.parameters(),\n",
    "                           lr=0.001)\n",
    "nl_module = SimpleModule.regression(nl_model,\n",
    "                                        optimizer=nl_optimizer,\n",
    "                                        metrics={'r2':R2Score()})\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2ac1a2df",
   "metadata": {},
   "source": [
    "We continue with the usual training steps, fit the model,\n",
    "and evaluate the test error. We see the test $R^2$ is a slight improvement over the linear AR model that also includes `day_of_week`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "id": "0591bf1d",
   "metadata": {
    "lines_to_next_cell": 0
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "GPU available: True (mps), used: True\n",
      "TPU available: False, using: 0 TPU cores\n",
      "IPU available: False, using: 0 IPUs\n",
      "HPU available: False, using: 0 HPUs\n",
      "\n",
      "  | Name  | Type             | Params\n",
      "-------------------------------------------\n",
      "0 | model | NonLinearARModel | 705   \n",
      "1 | loss  | MSELoss          | 0     \n",
      "-------------------------------------------\n",
      "705       Trainable params\n",
      "0         Non-trainable params\n",
      "705       Total params\n",
      "0.003     Total estimated model params size (MB)\n",
      "`Trainer.fit` stopped: `max_epochs=20` reached.\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\">        Test metric        </span>┃<span style=\"font-weight: bold\">       DataLoader 0        </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">         test_loss         </span>│<span style=\"color: #800080; text-decoration-color: #800080\">    0.5626183152198792     </span>│\n",
       "│<span style=\"color: #008080; text-decoration-color: #008080\">          test_r2          </span>│<span style=\"color: #800080; text-decoration-color: #800080\">    0.4660477638244629     </span>│\n",
       "└───────────────────────────┴───────────────────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1m       Test metric       \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      DataLoader 0       \u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━┩\n",
       "│\u001b[36m \u001b[0m\u001b[36m        test_loss        \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m   0.5626183152198792    \u001b[0m\u001b[35m \u001b[0m│\n",
       "│\u001b[36m \u001b[0m\u001b[36m         test_r2         \u001b[0m\u001b[36m \u001b[0m│\u001b[35m \u001b[0m\u001b[35m   0.4660477638244629    \u001b[0m\u001b[35m \u001b[0m│\n",
       "└───────────────────────────┴───────────────────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "[{'test_loss': 0.5626183152198792, 'test_r2': 0.4660477638244629}]"
      ]
     },
     "execution_count": 113,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "nl_trainer = Trainer(deterministic=True,\n",
    "                     max_epochs=20,\n",
    "                     enable_progress_bar=False,\n",
    "                     callbacks=[ErrorTracker()])\n",
    "nl_trainer.fit(nl_module, datamodule=day_dm)\n",
    "nl_trainer.test(nl_module, datamodule=day_dm) "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1bf8c38e",
   "metadata": {},
   "source": [
    "              \n",
    "\n",
    " \n"
   ]
  }
 ],
 "metadata": {
  "jupytext": {
   "cell_metadata_filter": "-all",
   "main_language": "python",
   "notebook_metadata_filter": "-all"
  },
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.12.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}