16_nlp_with_rnns_and_attention.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Chapter 16 – Natural Language Processing with RNNs and Attention**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "_This notebook contains all the sample code in chapter 16._"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<table align=\"left\">\n",
    "  <td>\n",
    "    <a target=\"_blank\" href=\"https://colab.research.google.com/github/ageron/handson-ml2/blob/master/16_nlp_with_rnns_and_attention.ipynb\"><img src=\"https://www.tensorflow.org/images/colab_logo_32px.png\" />Run in Google Colab</a>\n",
    "  </td>\n",
    "</table>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Setup"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First, let's import a few common modules, ensure MatplotLib plots figures inline and prepare a function to save the figures. We also check that Python 3.5 or later is installed (although Python 2.x may work, it is deprecated so we strongly recommend you use Python 3 instead), as well as Scikit-Learn ≥0.20 and TensorFlow ≥2.0."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Python ≥3.5 is required\n",
    "import sys\n",
    "assert sys.version_info >= (3, 5)\n",
    "\n",
    "# Scikit-Learn ≥0.20 is required\n",
    "import sklearn\n",
    "assert sklearn.__version__ >= \"0.20\"\n",
    "\n",
    "try:\n",
    "    # %tensorflow_version only exists in Colab.\n",
    "    %tensorflow_version 2.x\n",
    "    !pip install -q -U tensorflow-addons\n",
    "    IS_COLAB = True\n",
    "except Exception:\n",
    "    IS_COLAB = False\n",
    "\n",
    "# TensorFlow ≥2.0 is required\n",
    "import tensorflow as tf\n",
    "from tensorflow import keras\n",
    "assert tf.__version__ >= \"2.0\"\n",
    "\n",
    "if not tf.config.list_physical_devices('GPU'):\n",
    "    print(\"No GPU was detected. LSTMs and CNNs can be very slow without a GPU.\")\n",
    "    if IS_COLAB:\n",
    "        print(\"Go to Runtime > Change runtime and select a GPU hardware accelerator.\")\n",
    "\n",
    "# Common imports\n",
    "import numpy as np\n",
    "import os\n",
    "\n",
    "# to make this notebook's output stable across runs\n",
    "np.random.seed(42)\n",
    "tf.random.set_seed(42)\n",
    "\n",
    "# To plot pretty figures\n",
    "%matplotlib inline\n",
    "import matplotlib as mpl\n",
    "import matplotlib.pyplot as plt\n",
    "mpl.rc('axes', labelsize=14)\n",
    "mpl.rc('xtick', labelsize=12)\n",
    "mpl.rc('ytick', labelsize=12)\n",
    "\n",
    "# Where to save the figures\n",
    "PROJECT_ROOT_DIR = \".\"\n",
    "CHAPTER_ID = \"nlp\"\n",
    "IMAGES_PATH = os.path.join(PROJECT_ROOT_DIR, \"images\", CHAPTER_ID)\n",
    "os.makedirs(IMAGES_PATH, exist_ok=True)\n",
    "\n",
    "def save_fig(fig_id, tight_layout=True, fig_extension=\"png\", resolution=300):\n",
    "    path = os.path.join(IMAGES_PATH, fig_id + \".\" + fig_extension)\n",
    "    print(\"Saving figure\", fig_id)\n",
    "    if tight_layout:\n",
    "        plt.tight_layout()\n",
    "    plt.savefig(path, format=fig_extension, dpi=resolution)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Char-RNN"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Splitting a sequence into batches of shuffled windows"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For example, let's split the sequence 0 to 14 into windows of length 5, each shifted by 2 (e.g.,`[0, 1, 2, 3, 4]`, `[2, 3, 4, 5, 6]`, etc.), then shuffle them, and split them into inputs (the first 4 steps) and targets (the last 4 steps) (e.g., `[2, 3, 4, 5, 6]` would be split into `[[2, 3, 4, 5], [3, 4, 5, 6]]`), then create batches of 3 such input/target pairs:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "____________________ Batch 0 \n",
      "X_batch\n",
      "[[6 7 8 9]\n",
      " [2 3 4 5]\n",
      " [4 5 6 7]]\n",
      "===== \n",
      "Y_batch\n",
      "[[ 7  8  9 10]\n",
      " [ 3  4  5  6]\n",
      " [ 5  6  7  8]]\n",
      "____________________ Batch 1 \n",
      "X_batch\n",
      "[[ 0  1  2  3]\n",
      " [ 8  9 10 11]\n",
      " [10 11 12 13]]\n",
      "===== \n",
      "Y_batch\n",
      "[[ 1  2  3  4]\n",
      " [ 9 10 11 12]\n",
      " [11 12 13 14]]\n"
     ]
    }
   ],
   "source": [
    "np.random.seed(42)\n",
    "tf.random.set_seed(42)\n",
    "\n",
    "n_steps = 5\n",
    "dataset = tf.data.Dataset.from_tensor_slices(tf.range(15))\n",
    "dataset = dataset.window(n_steps, shift=2, drop_remainder=True)\n",
    "dataset = dataset.flat_map(lambda window: window.batch(n_steps))\n",
    "dataset = dataset.shuffle(10).map(lambda window: (window[:-1], window[1:]))\n",
    "dataset = dataset.batch(3).prefetch(1)\n",
    "for index, (X_batch, Y_batch) in enumerate(dataset):\n",
    "    print(\"_\" * 20, \"Batch\", index, \"\\nX_batch\")\n",
    "    print(X_batch.numpy())\n",
    "    print(\"=\" * 5, \"\\nY_batch\")\n",
    "    print(Y_batch.numpy())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Loading the Data and Preparing the Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading data from https://raw.githubusercontent.com/karpathy/char-rnn/master/data/tinyshakespeare/input.txt\n",
      "1122304/1115394 [==============================] - 0s 0us/step\n"
     ]
    }
   ],
   "source": [
    "shakespeare_url = \"https://raw.githubusercontent.com/karpathy/char-rnn/master/data/tinyshakespeare/input.txt\"\n",
    "filepath = keras.utils.get_file(\"shakespeare.txt\", shakespeare_url)\n",
    "with open(filepath) as f:\n",
    "    shakespeare_text = f.read()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "First Citizen:\n",
      "Before we proceed any further, hear me speak.\n",
      "\n",
      "All:\n",
      "Speak, speak.\n",
      "\n",
      "First Citizen:\n",
      "You are all resolved rather to die than to famish?\n",
      "\n"
     ]
    }
   ],
   "source": [
    "print(shakespeare_text[:148])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "\"\\n !$&',-.3:;?abcdefghijklmnopqrstuvwxyz\""
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "\"\".join(sorted(set(shakespeare_text.lower())))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "tokenizer = keras.preprocessing.text.Tokenizer(char_level=True)\n",
    "tokenizer.fit_on_texts(shakespeare_text)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[[20, 6, 9, 8, 3]]"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tokenizer.texts_to_sequences([\"First\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['f i r s t']"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tokenizer.sequences_to_texts([[20, 6, 9, 8, 3]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "max_id = len(tokenizer.word_index) # number of distinct characters\n",
    "dataset_size = tokenizer.document_count # total number of characters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "[encoded] = np.array(tokenizer.texts_to_sequences([shakespeare_text])) - 1\n",
    "train_size = dataset_size * 90 // 100\n",
    "dataset = tf.data.Dataset.from_tensor_slices(encoded[:train_size])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_steps = 100\n",
    "window_length = n_steps + 1 # target = input shifted 1 character ahead\n",
    "dataset = dataset.repeat().window(window_length, shift=1, drop_remainder=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = dataset.flat_map(lambda window: window.batch(window_length))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "np.random.seed(42)\n",
    "tf.random.set_seed(42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 32\n",
    "dataset = dataset.shuffle(10000).batch(batch_size)\n",
    "dataset = dataset.map(lambda windows: (windows[:, :-1], windows[:, 1:]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = dataset.map(\n",
    "    lambda X_batch, Y_batch: (tf.one_hot(X_batch, depth=max_id), Y_batch))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = dataset.prefetch(1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(32, 100, 39) (32, 100)\n"
     ]
    }
   ],
   "source": [
    "for X_batch, Y_batch in dataset.take(1):\n",
    "    print(X_batch.shape, Y_batch.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Creating and Training the Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train for 31370 steps\n",
      "Epoch 1/10\n",
      "31370/31370 [==============================] - 7150s 228ms/step - loss: 1.4671\n",
      "Epoch 2/10\n",
      "31370/31370 [==============================] - 7094s 226ms/step - loss: 1.3614\n",
      "Epoch 3/10\n",
      "31370/31370 [==============================] - 7063s 225ms/step - loss: 1.3404\n",
      "Epoch 4/10\n",
      "31370/31370 [==============================] - 7039s 224ms/step - loss: 1.3311\n",
      "Epoch 5/10\n",
      "31370/31370 [==============================] - 7056s 225ms/step - loss: 1.3256\n",
      "Epoch 6/10\n",
      "31370/31370 [==============================] - 7049s 225ms/step - loss: 1.3209\n",
      "Epoch 7/10\n",
      "31370/31370 [==============================] - 7068s 225ms/step - loss: 1.3166\n",
      "Epoch 8/10\n",
      "31370/31370 [==============================] - 7030s 224ms/step - loss: 1.3138\n",
      "Epoch 9/10\n",
      "31370/31370 [==============================] - 7061s 225ms/step - loss: 1.3120\n",
      "Epoch 10/10\n",
      "31370/31370 [==============================] - 7177s 229ms/step - loss: 1.3105\n"
     ]
    }
   ],
   "source": [
    "model = keras.models.Sequential([\n",
    "    keras.layers.GRU(128, return_sequences=True, input_shape=[None, max_id],\n",
    "                     dropout=0.2, recurrent_dropout=0.2),\n",
    "    keras.layers.GRU(128, return_sequences=True,\n",
    "                     dropout=0.2, recurrent_dropout=0.2),\n",
    "    keras.layers.TimeDistributed(keras.layers.Dense(max_id,\n",
    "                                                    activation=\"softmax\"))\n",
    "])\n",
    "model.compile(loss=\"sparse_categorical_crossentropy\", optimizer=\"adam\")\n",
    "history = model.fit(dataset, steps_per_epoch=train_size // batch_size,\n",
    "                    epochs=10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Using the Model to Generate Text"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "def preprocess(texts):\n",
    "    X = np.array(tokenizer.texts_to_sequences(texts)) - 1\n",
    "    return tf.one_hot(X, max_id)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'u'"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_new = preprocess([\"How are yo\"])\n",
    "Y_pred = model.predict_classes(X_new)\n",
    "tokenizer.sequences_to_texts(Y_pred + 1)[0][-1] # 1st sentence, last char"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0, 1, 0, 2, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 2, 1, 0, 2, 1,\n",
       "        0, 1, 2, 1, 1, 1, 2, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 2]])"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tf.random.set_seed(42)\n",
    "\n",
    "tf.random.categorical([[np.log(0.5), np.log(0.4), np.log(0.1)]], num_samples=40).numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "def next_char(text, temperature=1):\n",
    "    X_new = preprocess([text])\n",
    "    y_proba = model.predict(X_new)[0, -1:, :]\n",
    "    rescaled_logits = tf.math.log(y_proba) / temperature\n",
    "    char_id = tf.random.categorical(rescaled_logits, num_samples=1) + 1\n",
    "    return tokenizer.sequences_to_texts(char_id.numpy())[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'u'"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tf.random.set_seed(42)\n",
    "\n",
    "next_char(\"How are yo\", temperature=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "def complete_text(text, n_chars=50, temperature=1):\n",
    "    for _ in range(n_chars):\n",
    "        text += next_char(text, temperature)\n",
    "    return text"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:5 out of the last 6 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:6 out of the last 7 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:7 out of the last 8 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:8 out of the last 9 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:9 out of the last 10 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:10 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:10 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:10 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:10 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:10 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:10 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:10 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:10 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:10 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:10 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 12 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44616830> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "the belly the charges of the other words\n",
      "and belly \n"
     ]
    }
   ],
   "source": [
    "tf.random.set_seed(42)\n",
    "\n",
    "print(complete_text(\"t\", temperature=0.2))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "thing! they know't.\n",
      "\n",
      "biondello:\n",
      "for you are the own\n"
     ]
    }
   ],
   "source": [
    "print(complete_text(\"t\", temperature=1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "th no cyty\n",
      "use ffor was firive this toighingaber; b\n"
     ]
    }
   ],
   "source": [
    "print(complete_text(\"t\", temperature=2))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Stateful RNN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "tf.random.set_seed(42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = tf.data.Dataset.from_tensor_slices(encoded[:train_size])\n",
    "dataset = dataset.window(window_length, shift=n_steps, drop_remainder=True)\n",
    "dataset = dataset.flat_map(lambda window: window.batch(window_length))\n",
    "dataset = dataset.repeat().batch(1)\n",
    "dataset = dataset.map(lambda windows: (windows[:, :-1], windows[:, 1:]))\n",
    "dataset = dataset.map(\n",
    "    lambda X_batch, Y_batch: (tf.one_hot(X_batch, depth=max_id), Y_batch))\n",
    "dataset = dataset.prefetch(1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 32\n",
    "encoded_parts = np.array_split(encoded[:train_size], batch_size)\n",
    "datasets = []\n",
    "for encoded_part in encoded_parts:\n",
    "    dataset = tf.data.Dataset.from_tensor_slices(encoded_part)\n",
    "    dataset = dataset.window(window_length, shift=n_steps, drop_remainder=True)\n",
    "    dataset = dataset.flat_map(lambda window: window.batch(window_length))\n",
    "    datasets.append(dataset)\n",
    "dataset = tf.data.Dataset.zip(tuple(datasets)).map(lambda *windows: tf.stack(windows))\n",
    "dataset = dataset.repeat().map(lambda windows: (windows[:, :-1], windows[:, 1:]))\n",
    "dataset = dataset.map(\n",
    "    lambda X_batch, Y_batch: (tf.one_hot(X_batch, depth=max_id), Y_batch))\n",
    "dataset = dataset.prefetch(1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = keras.models.Sequential([\n",
    "    keras.layers.GRU(128, return_sequences=True, stateful=True,\n",
    "                     dropout=0.2, recurrent_dropout=0.2,\n",
    "                     batch_input_shape=[batch_size, None, max_id]),\n",
    "    keras.layers.GRU(128, return_sequences=True, stateful=True,\n",
    "                     dropout=0.2, recurrent_dropout=0.2),\n",
    "    keras.layers.TimeDistributed(keras.layers.Dense(max_id,\n",
    "                                                    activation=\"softmax\"))\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ResetStatesCallback(keras.callbacks.Callback):\n",
    "    def on_epoch_begin(self, epoch, logs):\n",
    "        self.model.reset_states()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train for 313 steps\n",
      "Epoch 1/50\n",
      "313/313 [==============================] - 62s 198ms/step - loss: 2.6189\n",
      "Epoch 2/50\n",
      "313/313 [==============================] - 58s 187ms/step - loss: 2.2091\n",
      "Epoch 3/50\n",
      "313/313 [==============================] - 56s 178ms/step - loss: 2.0775\n",
      "Epoch 4/50\n",
      "313/313 [==============================] - 56s 179ms/step - loss: 2.4689\n",
      "Epoch 5/50\n",
      "313/313 [==============================] - 56s 179ms/step - loss: 2.3274\n",
      "Epoch 6/50\n",
      "313/313 [==============================] - 57s 183ms/step - loss: 2.1412\n",
      "Epoch 7/50\n",
      "313/313 [==============================] - 57s 183ms/step - loss: 2.0748\n",
      "Epoch 8/50\n",
      "313/313 [==============================] - 56s 179ms/step - loss: 1.9850\n",
      "Epoch 9/50\n",
      "313/313 [==============================] - 56s 179ms/step - loss: 1.9465\n",
      "Epoch 10/50\n",
      "313/313 [==============================] - 56s 179ms/step - loss: 1.8995\n",
      "Epoch 11/50\n",
      "313/313 [==============================] - 57s 182ms/step - loss: 1.8576\n",
      "Epoch 12/50\n",
      "313/313 [==============================] - 56s 179ms/step - loss: 1.8510\n",
      "Epoch 13/50\n",
      "313/313 [==============================] - 57s 184ms/step - loss: 1.8038\n",
      "Epoch 14/50\n",
      "313/313 [==============================] - 56s 178ms/step - loss: 1.7867\n",
      "Epoch 15/50\n",
      "313/313 [==============================] - 56s 180ms/step - loss: 1.7635\n",
      "Epoch 16/50\n",
      "313/313 [==============================] - 56s 179ms/step - loss: 1.7270\n",
      "Epoch 17/50\n",
      "313/313 [==============================] - 58s 184ms/step - loss: 1.7097\n",
      "Epoch 18/50\n",
      "313/313 [==============================] - 58s 186ms/step - loss: 1.6970\n",
      "Epoch 19/50\n",
      "313/313 [==============================] - 57s 182ms/step - loss: 1.6853\n",
      "Epoch 20/50\n",
      "313/313 [==============================] - 56s 179ms/step - loss: 1.6739\n",
      "Epoch 21/50\n",
      "313/313 [==============================] - 56s 179ms/step - loss: 1.6641\n",
      "Epoch 22/50\n",
      "313/313 [==============================] - 57s 182ms/step - loss: 1.6567\n",
      "Epoch 23/50\n",
      "313/313 [==============================] - 56s 179ms/step - loss: 1.6473\n",
      "Epoch 24/50\n",
      "313/313 [==============================] - 57s 183ms/step - loss: 1.6417\n",
      "Epoch 25/50\n",
      "313/313 [==============================] - 56s 179ms/step - loss: 1.6356\n",
      "Epoch 26/50\n",
      "313/313 [==============================] - 57s 182ms/step - loss: 1.6296\n",
      "Epoch 27/50\n",
      "313/313 [==============================] - 57s 181ms/step - loss: 1.6258\n",
      "Epoch 28/50\n",
      "313/313 [==============================] - 58s 184ms/step - loss: 1.6193\n",
      "Epoch 29/50\n",
      "313/313 [==============================] - 58s 186ms/step - loss: 1.6155\n",
      "Epoch 30/50\n",
      "313/313 [==============================] - 57s 182ms/step - loss: 1.6103\n",
      "Epoch 31/50\n",
      "313/313 [==============================] - 57s 182ms/step - loss: 1.6071\n",
      "Epoch 32/50\n",
      "313/313 [==============================] - 57s 182ms/step - loss: 1.6018\n",
      "Epoch 33/50\n",
      "313/313 [==============================] - 58s 185ms/step - loss: 1.5998\n",
      "Epoch 34/50\n",
      "313/313 [==============================] - 58s 184ms/step - loss: 1.5954\n",
      "Epoch 35/50\n",
      "313/313 [==============================] - 58s 185ms/step - loss: 1.5944\n",
      "Epoch 36/50\n",
      "313/313 [==============================] - 57s 183ms/step - loss: 1.5902\n",
      "Epoch 37/50\n",
      "313/313 [==============================] - 57s 183ms/step - loss: 1.5893\n",
      "Epoch 38/50\n",
      "313/313 [==============================] - 59s 187ms/step - loss: 1.5845\n",
      "Epoch 39/50\n",
      "313/313 [==============================] - 57s 183ms/step - loss: 1.5821\n",
      "Epoch 40/50\n",
      "313/313 [==============================] - 59s 187ms/step - loss: 1.5798\n",
      "Epoch 41/50\n",
      "313/313 [==============================] - 57s 181ms/step - loss: 1.5794\n",
      "Epoch 42/50\n",
      "313/313 [==============================] - 57s 182ms/step - loss: 1.5774\n",
      "Epoch 43/50\n",
      "313/313 [==============================] - 57s 182ms/step - loss: 1.5755\n",
      "Epoch 44/50\n",
      "313/313 [==============================] - 58s 186ms/step - loss: 1.5735\n",
      "Epoch 45/50\n",
      "313/313 [==============================] - 58s 186ms/step - loss: 1.5714\n",
      "Epoch 46/50\n",
      "313/313 [==============================] - 57s 181ms/step - loss: 1.5686\n",
      "Epoch 47/50\n",
      "313/313 [==============================] - 57s 181ms/step - loss: 1.5675\n",
      "Epoch 48/50\n",
      "313/313 [==============================] - 56s 180ms/step - loss: 1.5657\n",
      "Epoch 49/50\n",
      "313/313 [==============================] - 58s 185ms/step - loss: 1.5654\n",
      "Epoch 50/50\n",
      "313/313 [==============================] - 57s 182ms/step - loss: 1.5620\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7f8d45d95d10>"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.compile(loss=\"sparse_categorical_crossentropy\", optimizer=\"adam\")\n",
    "steps_per_epoch = train_size // batch_size // n_steps\n",
    "model.fit(dataset, steps_per_epoch=steps_per_epoch, epochs=50,\n",
    "                   callbacks=[ResetStatesCallback()])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To use the model with different batch sizes, we need to create a stateless copy. We can get rid of dropout since it is only used during training:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "stateless_model = keras.models.Sequential([\n",
    "    keras.layers.GRU(128, return_sequences=True, input_shape=[None, max_id]),\n",
    "    keras.layers.GRU(128, return_sequences=True),\n",
    "    keras.layers.TimeDistributed(keras.layers.Dense(max_id,\n",
    "                                                    activation=\"softmax\"))\n",
    "])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To set the weights, we first need to build the model (so the weights get created):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "stateless_model.build(tf.TensorShape([None, None, max_id]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "stateless_model.set_weights(model.get_weights())\n",
    "model = stateless_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:5 out of the last 5 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:6 out of the last 6 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:7 out of the last 7 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:8 out of the last 8 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:9 out of the last 9 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:10 out of the last 10 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "WARNING:tensorflow:11 out of the last 11 calls to <function _make_execution_function.<locals>.distributed_function at 0x7f8d44bc53b0> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings is likely due to passing python objects instead of tensors. Also, tf.function has experimental_relax_shapes=True option that relaxes argument shapes that can avoid unnecessary retracing. Please refer to https://www.tensorflow.org/tutorials/customization/performance#python_or_tensor_args and https://www.tensorflow.org/api_docs/python/tf/function for more details.\n",
      "tor:\n",
      "in the negver up how it thou like him;\n",
      "when it\n"
     ]
    }
   ],
   "source": [
    "tf.random.set_seed(42)\n",
    "\n",
    "print(complete_text(\"t\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Sentiment Analysis"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "tf.random.set_seed(42)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "You can load the IMDB dataset easily:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/imdb.npz\n",
      "17465344/17464789 [==============================] - 0s 0us/step\n"
     ]
    }
   ],
   "source": [
    "(X_train, y_test), (X_valid, y_test) = keras.datasets.imdb.load_data()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[1, 14, 22, 16, 43, 530, 973, 1622, 1385, 65]"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train[0][:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/imdb_word_index.json\n",
      "1646592/1641221 [==============================] - 0s 0us/step\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "'<sos> this film was just brilliant casting location scenery story'"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "word_index = keras.datasets.imdb.get_word_index()\n",
    "id_to_word = {id_ + 3: word for word, id_ in word_index.items()}\n",
    "for id_, token in enumerate((\"<pad>\", \"<sos>\", \"<unk>\")):\n",
    "    id_to_word[id_] = token\n",
    "\" \".join([id_to_word[id_] for id_ in X_train[0][:10]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1mDownloading and preparing dataset imdb_reviews/plain_text/1.0.0 (download: 80.23 MiB, generated: Unknown size, total: 80.23 MiB) to /home/aurelien_geron_kiwisoft_io/tensorflow_datasets/imdb_reviews/plain_text/1.0.0...\u001b[0m\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "9d19e9f5f622440b9feb5ed1a98db7b3",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=1.0, bar_style='info', description='Dl Completed...', max=1.0, style=Progre…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "001634a67f0e42f69450a95ed84492b2",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=1.0, bar_style='info', description='Dl Size...', max=1.0, style=ProgressSty…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\n",
      "\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=1.0, bar_style='info', max=1.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shuffling and writing examples to /home/aurelien_geron_kiwisoft_io/tensorflow_datasets/imdb_reviews/plain_text/1.0.0.incompleteK5RNB1/imdb_reviews-train.tfrecord\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "c661939803ee4442be234cc9e1485599",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=25000.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=1.0, bar_style='info', max=1.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shuffling and writing examples to /home/aurelien_geron_kiwisoft_io/tensorflow_datasets/imdb_reviews/plain_text/1.0.0.incompleteK5RNB1/imdb_reviews-test.tfrecord\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "bdd0d660b82e4f93b4824e6e3573f197",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=25000.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=1.0, bar_style='info', max=1.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shuffling and writing examples to /home/aurelien_geron_kiwisoft_io/tensorflow_datasets/imdb_reviews/plain_text/1.0.0.incompleteK5RNB1/imdb_reviews-unsupervised.tfrecord\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "91133fb7161d440383d4d50bd06f16ba",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(FloatProgress(value=0.0, max=50000.0), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1mDataset imdb_reviews downloaded and prepared to /home/aurelien_geron_kiwisoft_io/tensorflow_datasets/imdb_reviews/plain_text/1.0.0. Subsequent calls will reuse this data.\u001b[0m\n",
      "\r"
     ]
    }
   ],
   "source": [
    "import tensorflow_datasets as tfds\n",
    "\n",
    "datasets, info = tfds.load(\"imdb_reviews\", as_supervised=True, with_info=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "dict_keys(['test', 'train', 'unsupervised'])"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "datasets.keys()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_size = info.splits[\"train\"].num_examples\n",
    "test_size = info.splits[\"test\"].num_examples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(25000, 25000)"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_size, test_size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Review: This was an absolutely terrible movie. Don't be lured in by Christopher Walken or Michael Ironside. Both are great actors, but this must simply be their worst role in history. Even their great acting  ...\n",
      "Label: 0 = Negative\n",
      "\n",
      "Review: I have been known to fall asleep during films, but this is usually due to a combination of things including, really tired, being warm and comfortable on the sette and having just eaten a lot. However  ...\n",
      "Label: 0 = Negative\n",
      "\n"
     ]
    }
   ],
   "source": [
    "for X_batch, y_batch in datasets[\"train\"].batch(2).take(1):\n",
    "    for review, label in zip(X_batch.numpy(), y_batch.numpy()):\n",
    "        print(\"Review:\", review.decode(\"utf-8\")[:200], \"...\")\n",
    "        print(\"Label:\", label, \"= Positive\" if label else \"= Negative\")\n",
    "        print()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "def preprocess(X_batch, y_batch):\n",
    "    X_batch = tf.strings.substr(X_batch, 0, 300)\n",
    "    X_batch = tf.strings.regex_replace(X_batch, rb\"<br\\s*/?>\", b\" \")\n",
    "    X_batch = tf.strings.regex_replace(X_batch, b\"[^a-zA-Z']\", b\" \")\n",
    "    X_batch = tf.strings.split(X_batch)\n",
    "    return X_batch.to_tensor(default_value=b\"<pad>\"), y_batch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(<tf.Tensor: shape=(2, 53), dtype=string, numpy=\n",
       " array([[b'This', b'was', b'an', b'absolutely', b'terrible', b'movie',\n",
       "         b\"Don't\", b'be', b'lured', b'in', b'by', b'Christopher',\n",
       "         b'Walken', b'or', b'Michael', b'Ironside', b'Both', b'are',\n",
       "         b'great', b'actors', b'but', b'this', b'must', b'simply', b'be',\n",
       "         b'their', b'worst', b'role', b'in', b'history', b'Even',\n",
       "         b'their', b'great', b'acting', b'could', b'not', b'redeem',\n",
       "         b'this', b\"movie's\", b'ridiculous', b'storyline', b'This',\n",
       "         b'movie', b'is', b'an', b'early', b'nineties', b'US',\n",
       "         b'propaganda', b'pi', b'<pad>', b'<pad>', b'<pad>'],\n",
       "        [b'I', b'have', b'been', b'known', b'to', b'fall', b'asleep',\n",
       "         b'during', b'films', b'but', b'this', b'is', b'usually', b'due',\n",
       "         b'to', b'a', b'combination', b'of', b'things', b'including',\n",
       "         b'really', b'tired', b'being', b'warm', b'and', b'comfortable',\n",
       "         b'on', b'the', b'sette', b'and', b'having', b'just', b'eaten',\n",
       "         b'a', b'lot', b'However', b'on', b'this', b'occasion', b'I',\n",
       "         b'fell', b'asleep', b'because', b'the', b'film', b'was',\n",
       "         b'rubbish', b'The', b'plot', b'development', b'was', b'constant',\n",
       "         b'Cons']], dtype=object)>,\n",
       " <tf.Tensor: shape=(2,), dtype=int64, numpy=array([0, 0])>)"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "preprocess(X_batch, y_batch)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "from collections import Counter\n",
    "\n",
    "vocabulary = Counter()\n",
    "for X_batch, y_batch in datasets[\"train\"].batch(32).map(preprocess):\n",
    "    for review in X_batch:\n",
    "        vocabulary.update(list(review.numpy()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[(b'<pad>', 214309), (b'the', 61137), (b'a', 38564)]"
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "vocabulary.most_common()[:3]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "53893"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(vocabulary)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "vocab_size = 10000\n",
    "truncated_vocabulary = [\n",
    "    word for word, count in vocabulary.most_common()[:vocab_size]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "22\n",
      "12\n",
      "11\n",
      "10000\n"
     ]
    }
   ],
   "source": [
    "word_to_id = {word: index for index, word in enumerate(truncated_vocabulary)}\n",
    "for word in b\"This movie was faaaaaantastic\".split():\n",
    "    print(word_to_id.get(word) or vocab_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [],
   "source": [
    "words = tf.constant(truncated_vocabulary)\n",
    "word_ids = tf.range(len(truncated_vocabulary), dtype=tf.int64)\n",
    "vocab_init = tf.lookup.KeyValueTensorInitializer(words, word_ids)\n",
    "num_oov_buckets = 1000\n",
    "table = tf.lookup.StaticVocabularyTable(vocab_init, num_oov_buckets)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: shape=(1, 4), dtype=int64, numpy=array([[   22,    12,    11, 10053]])>"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "table.lookup(tf.constant([b\"This movie was faaaaaantastic\".split()]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "def encode_words(X_batch, y_batch):\n",
    "    return table.lookup(X_batch), y_batch\n",
    "\n",
    "train_set = datasets[\"train\"].repeat().batch(32).map(preprocess)\n",
    "train_set = train_set.map(encode_words).prefetch(1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tf.Tensor(\n",
      "[[  22   11   28 ...    0    0    0]\n",
      " [   6   21   70 ...    0    0    0]\n",
      " [4099 6881    1 ...    0    0    0]\n",
      " ...\n",
      " [  22   12  118 ...  331 1047    0]\n",
      " [1757 4101  451 ...    0    0    0]\n",
      " [3365 4392    6 ...    0    0    0]], shape=(32, 60), dtype=int64)\n",
      "tf.Tensor([0 0 0 1 1 1 0 0 0 0 0 1 1 0 1 0 1 1 1 0 1 1 1 1 1 0 0 0 1 0 0 0], shape=(32,), dtype=int64)\n"
     ]
    }
   ],
   "source": [
    "for X_batch, y_batch in train_set.take(1):\n",
    "    print(X_batch)\n",
    "    print(y_batch)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train for 781 steps\n",
      "Epoch 1/5\n",
      "781/781 [==============================] - 118s 152ms/step - loss: 0.5305 - accuracy: 0.7282\n",
      "Epoch 2/5\n",
      "781/781 [==============================] - 113s 145ms/step - loss: 0.3459 - accuracy: 0.8554\n",
      "Epoch 3/5\n",
      "781/781 [==============================] - 113s 145ms/step - loss: 0.1913 - accuracy: 0.9319\n",
      "Epoch 4/5\n",
      "781/781 [==============================] - 114s 146ms/step - loss: 0.1341 - accuracy: 0.9535\n",
      "Epoch 5/5\n",
      "781/781 [==============================] - 116s 148ms/step - loss: 0.1011 - accuracy: 0.9624\n"
     ]
    }
   ],
   "source": [
    "embed_size = 128\n",
    "model = keras.models.Sequential([\n",
    "    keras.layers.Embedding(vocab_size + num_oov_buckets, embed_size,\n",
    "                           mask_zero=True, # not shown in the book\n",
    "                           input_shape=[None]),\n",
    "    keras.layers.GRU(128, return_sequences=True),\n",
    "    keras.layers.GRU(128),\n",
    "    keras.layers.Dense(1, activation=\"sigmoid\")\n",
    "])\n",
    "model.compile(loss=\"binary_crossentropy\", optimizer=\"adam\", metrics=[\"accuracy\"])\n",
    "history = model.fit(train_set, steps_per_epoch=train_size // 32, epochs=5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Or using manual masking:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train for 781 steps\n",
      "Epoch 1/5\n",
      "781/781 [==============================] - 118s 152ms/step - loss: 0.5425 - accuracy: 0.7155\n",
      "Epoch 2/5\n",
      "781/781 [==============================] - 112s 143ms/step - loss: 0.3479 - accuracy: 0.8558\n",
      "Epoch 3/5\n",
      "781/781 [==============================] - 112s 144ms/step - loss: 0.1761 - accuracy: 0.9388\n",
      "Epoch 4/5\n",
      "781/781 [==============================] - 115s 147ms/step - loss: 0.1281 - accuracy: 0.9531\n",
      "Epoch 5/5\n",
      "781/781 [==============================] - 116s 148ms/step - loss: 0.1088 - accuracy: 0.9603\n"
     ]
    }
   ],
   "source": [
    "K = keras.backend\n",
    "embed_size = 128\n",
    "inputs = keras.layers.Input(shape=[None])\n",
    "mask = keras.layers.Lambda(lambda inputs: K.not_equal(inputs, 0))(inputs)\n",
    "z = keras.layers.Embedding(vocab_size + num_oov_buckets, embed_size)(inputs)\n",
    "z = keras.layers.GRU(128, return_sequences=True)(z, mask=mask)\n",
    "z = keras.layers.GRU(128)(z, mask=mask)\n",
    "outputs = keras.layers.Dense(1, activation=\"sigmoid\")(z)\n",
    "model = keras.models.Model(inputs=[inputs], outputs=[outputs])\n",
    "model.compile(loss=\"binary_crossentropy\", optimizer=\"adam\", metrics=[\"accuracy\"])\n",
    "history = model.fit(train_set, steps_per_epoch=train_size // 32, epochs=5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Reusing Pretrained Embeddings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "tf.random.set_seed(42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [],
   "source": [
    "TFHUB_CACHE_DIR = os.path.join(os.curdir, \"my_tfhub_cache\")\n",
    "os.environ[\"TFHUB_CACHE_DIR\"] = TFHUB_CACHE_DIR"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow_hub as hub\n",
    "\n",
    "model = keras.Sequential([\n",
    "    hub.KerasLayer(\"https://tfhub.dev/google/tf2-preview/nnlm-en-dim50/1\",\n",
    "                   dtype=tf.string, input_shape=[], output_shape=[50]),\n",
    "    keras.layers.Dense(128, activation=\"relu\"),\n",
    "    keras.layers.Dense(1, activation=\"sigmoid\")\n",
    "])\n",
    "model.compile(loss=\"binary_crossentropy\", optimizer=\"adam\",\n",
    "              metrics=[\"accuracy\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "./my_tfhub_cache/82c4aaf4250ffb09088bd48368ee7fd00e5464fe.descriptor.txt\n",
      "./my_tfhub_cache/82c4aaf4250ffb09088bd48368ee7fd00e5464fe/saved_model.pb\n",
      "./my_tfhub_cache/82c4aaf4250ffb09088bd48368ee7fd00e5464fe/variables/variables.data-00000-of-00001\n",
      "./my_tfhub_cache/82c4aaf4250ffb09088bd48368ee7fd00e5464fe/variables/variables.index\n",
      "./my_tfhub_cache/82c4aaf4250ffb09088bd48368ee7fd00e5464fe/assets/tokens.txt\n"
     ]
    }
   ],
   "source": [
    "for dirpath, dirnames, filenames in os.walk(TFHUB_CACHE_DIR):\n",
    "    for filename in filenames:\n",
    "        print(os.path.join(dirpath, filename))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train for 781 steps\n",
      "Epoch 1/5\n",
      "781/781 [==============================] - 128s 164ms/step - loss: 0.5460 - accuracy: 0.7267\n",
      "Epoch 2/5\n",
      "781/781 [==============================] - 128s 164ms/step - loss: 0.5129 - accuracy: 0.7495\n",
      "Epoch 3/5\n",
      "781/781 [==============================] - 129s 165ms/step - loss: 0.5082 - accuracy: 0.7530\n",
      "Epoch 4/5\n",
      "781/781 [==============================] - 128s 164ms/step - loss: 0.5047 - accuracy: 0.7533\n",
      "Epoch 5/5\n",
      "781/781 [==============================] - 128s 164ms/step - loss: 0.5015 - accuracy: 0.7560\n"
     ]
    }
   ],
   "source": [
    "import tensorflow_datasets as tfds\n",
    "\n",
    "datasets, info = tfds.load(\"imdb_reviews\", as_supervised=True, with_info=True)\n",
    "train_size = info.splits[\"train\"].num_examples\n",
    "batch_size = 32\n",
    "train_set = datasets[\"train\"].repeat().batch(batch_size).prefetch(1)\n",
    "history = model.fit(train_set, steps_per_epoch=train_size // batch_size, epochs=5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Automatic Translation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [],
   "source": [
    "tf.random.set_seed(42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "vocab_size = 100\n",
    "embed_size = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow_addons as tfa\n",
    "\n",
    "encoder_inputs = keras.layers.Input(shape=[None], dtype=np.int32)\n",
    "decoder_inputs = keras.layers.Input(shape=[None], dtype=np.int32)\n",
    "sequence_lengths = keras.layers.Input(shape=[], dtype=np.int32)\n",
    "\n",
    "embeddings = keras.layers.Embedding(vocab_size, embed_size)\n",
    "encoder_embeddings = embeddings(encoder_inputs)\n",
    "decoder_embeddings = embeddings(decoder_inputs)\n",
    "\n",
    "encoder = keras.layers.LSTM(512, return_state=True)\n",
    "encoder_outputs, state_h, state_c = encoder(encoder_embeddings)\n",
    "encoder_state = [state_h, state_c]\n",
    "\n",
    "sampler = tfa.seq2seq.sampler.TrainingSampler()\n",
    "\n",
    "decoder_cell = keras.layers.LSTMCell(512)\n",
    "output_layer = keras.layers.Dense(vocab_size)\n",
    "decoder = tfa.seq2seq.basic_decoder.BasicDecoder(decoder_cell, sampler,\n",
    "                                                 output_layer=output_layer)\n",
    "final_outputs, final_state, final_sequence_lengths = decoder(\n",
    "    decoder_embeddings, initial_state=encoder_state,\n",
    "    sequence_length=sequence_lengths)\n",
    "Y_proba = tf.nn.softmax(final_outputs.rnn_output)\n",
    "\n",
    "model = keras.models.Model(\n",
    "    inputs=[encoder_inputs, decoder_inputs, sequence_lengths],\n",
    "    outputs=[Y_proba])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss=\"sparse_categorical_crossentropy\", optimizer=\"adam\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 1000 samples\n",
      "Epoch 1/2\n",
      "1000/1000 [==============================] - 6s 6ms/sample - loss: 4.6053\n",
      "Epoch 2/2\n",
      "1000/1000 [==============================] - 3s 3ms/sample - loss: 4.6031\n"
     ]
    }
   ],
   "source": [
    "X = np.random.randint(100, size=10*1000).reshape(1000, 10)\n",
    "Y = np.random.randint(100, size=15*1000).reshape(1000, 15)\n",
    "X_decoder = np.c_[np.zeros((1000, 1)), Y[:, :-1]]\n",
    "seq_lengths = np.full([1000], 15)\n",
    "\n",
    "history = model.fit([X, X_decoder, seq_lengths], Y, epochs=2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Bidirectional Recurrent Layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_5\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "gru_10 (GRU)                 (None, None, 10)          660       \n",
      "_________________________________________________________________\n",
      "bidirectional (Bidirectional (None, None, 20)          1320      \n",
      "=================================================================\n",
      "Total params: 1,980\n",
      "Trainable params: 1,980\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = keras.models.Sequential([\n",
    "    keras.layers.GRU(10, return_sequences=True, input_shape=[None, 10]),\n",
    "    keras.layers.Bidirectional(keras.layers.GRU(10, return_sequences=True))\n",
    "])\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Positional Encoding"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [],
   "source": [
    "class PositionalEncoding(keras.layers.Layer):\n",
    "    def __init__(self, max_steps, max_dims, dtype=tf.float32, **kwargs):\n",
    "        super().__init__(dtype=dtype, **kwargs)\n",
    "        if max_dims % 2 == 1: max_dims += 1 # max_dims must be even\n",
    "        p, i = np.meshgrid(np.arange(max_steps), np.arange(max_dims // 2))\n",
    "        pos_emb = np.empty((1, max_steps, max_dims))\n",
    "        pos_emb[0, :, ::2] = np.sin(p / 10000**(2 * i / max_dims)).T\n",
    "        pos_emb[0, :, 1::2] = np.cos(p / 10000**(2 * i / max_dims)).T\n",
    "        self.positional_embedding = tf.constant(pos_emb.astype(self.dtype))\n",
    "    def call(self, inputs):\n",
    "        shape = tf.shape(inputs)\n",
    "        return inputs + self.positional_embedding[:, :shape[-2], :shape[-1]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [],
   "source": [
    "max_steps = 201\n",
    "max_dims = 512\n",
    "pos_emb = PositionalEncoding(max_steps, max_dims)\n",
    "PE = pos_emb(np.zeros((1, max_steps, max_dims), np.float32))[0].numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "i1, i2, crop_i = 100, 101, 150\n",
    "p1, p2, p3 = 22, 60, 35\n",
    "fig, (ax1, ax2) = plt.subplots(nrows=2, ncols=1, sharex=True, figsize=(9, 5))\n",
    "ax1.plot([p1, p1], [-1, 1], \"k--\", label=\"$p = {}$\".format(p1))\n",
    "ax1.plot([p2, p2], [-1, 1], \"k--\", label=\"$p = {}$\".format(p2), alpha=0.5)\n",
    "ax1.plot(p3, PE[p3, i1], \"bx\", label=\"$p = {}$\".format(p3))\n",
    "ax1.plot(PE[:,i1], \"b-\", label=\"$i = {}$\".format(i1))\n",
    "ax1.plot(PE[:,i2], \"r-\", label=\"$i = {}$\".format(i2))\n",
    "ax1.plot([p1, p2], [PE[p1, i1], PE[p2, i1]], \"bo\")\n",
    "ax1.plot([p1, p2], [PE[p1, i2], PE[p2, i2]], \"ro\")\n",
    "ax1.legend(loc=\"center right\", fontsize=14, framealpha=0.95)\n",
    "ax1.set_ylabel(\"$P_{(p,i)}$\", rotation=0, fontsize=16)\n",
    "ax1.grid(True, alpha=0.3)\n",
    "ax1.hlines(0, 0, max_steps - 1, color=\"k\", linewidth=1, alpha=0.3)\n",
    "ax1.axis([0, max_steps - 1, -1, 1])\n",
    "ax2.imshow(PE.T[:crop_i], cmap=\"gray\", interpolation=\"bilinear\", aspect=\"auto\")\n",
    "ax2.hlines(i1, 0, max_steps - 1, color=\"b\")\n",
    "cheat = 2 # need to raise the red line a bit, or else it hides the blue one\n",
    "ax2.hlines(i2+cheat, 0, max_steps - 1, color=\"r\")\n",
    "ax2.plot([p1, p1], [0, crop_i], \"k--\")\n",
    "ax2.plot([p2, p2], [0, crop_i], \"k--\", alpha=0.5)\n",
    "ax2.plot([p1, p2], [i2+cheat, i2+cheat], \"ro\")\n",
    "ax2.plot([p1, p2], [i1, i1], \"bo\")\n",
    "ax2.axis([0, max_steps - 1, 0, crop_i])\n",
    "ax2.set_xlabel(\"$p$\", fontsize=16)\n",
    "ax2.set_ylabel(\"$i$\", rotation=0, fontsize=16)\n",
    "plt.savefig(\"positional_embedding_plot\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [],
   "source": [
    "embed_size = 512; max_steps = 500; vocab_size = 10000\n",
    "encoder_inputs = keras.layers.Input(shape=[None], dtype=np.int32)\n",
    "decoder_inputs = keras.layers.Input(shape=[None], dtype=np.int32)\n",
    "embeddings = keras.layers.Embedding(vocab_size, embed_size)\n",
    "encoder_embeddings = embeddings(encoder_inputs)\n",
    "decoder_embeddings = embeddings(decoder_inputs)\n",
    "positional_encoding = PositionalEncoding(max_steps, max_dims=embed_size)\n",
    "encoder_in = positional_encoding(encoder_embeddings)\n",
    "decoder_in = positional_encoding(decoder_embeddings)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here is a (very) simplified Transformer (the actual architecture has skip connections, layer norm, dense nets, and most importantly it uses Multi-Head Attention instead of regular Attention):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [],
   "source": [
    "Z = encoder_in\n",
    "for N in range(6):\n",
    "    Z = keras.layers.Attention(use_scale=True)([Z, Z])\n",
    "\n",
    "encoder_outputs = Z\n",
    "Z = decoder_in\n",
    "for N in range(6):\n",
    "    Z = keras.layers.Attention(use_scale=True, causal=True)([Z, Z])\n",
    "    Z = keras.layers.Attention(use_scale=True)([Z, encoder_outputs])\n",
    "\n",
    "outputs = keras.layers.TimeDistributed(\n",
    "    keras.layers.Dense(vocab_size, activation=\"softmax\"))(Z)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here's a basic implementation of the `MultiHeadAttention` layer. One will likely be added to `keras.layers` in the near future. Note that `Conv1D` layers with `kernel_size=1` (and the default `padding=\"valid\"` and `strides=1`) is equivalent to a `TimeDistributed(Dense(...))` layer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [],
   "source": [
    "K = keras.backend\n",
    "\n",
    "class MultiHeadAttention(keras.layers.Layer):\n",
    "    def __init__(self, n_heads, causal=False, use_scale=False, **kwargs):\n",
    "        self.n_heads = n_heads\n",
    "        self.causal = causal\n",
    "        self.use_scale = use_scale\n",
    "        super().__init__(**kwargs)\n",
    "    def build(self, batch_input_shape):\n",
    "        self.dims = batch_input_shape[0][-1]\n",
    "        self.q_dims, self.v_dims, self.k_dims = [self.dims // self.n_heads] * 3 # could be hyperparameters instead\n",
    "        self.q_linear = keras.layers.Conv1D(self.n_heads * self.q_dims, kernel_size=1, use_bias=False)\n",
    "        self.v_linear = keras.layers.Conv1D(self.n_heads * self.v_dims, kernel_size=1, use_bias=False)\n",
    "        self.k_linear = keras.layers.Conv1D(self.n_heads * self.k_dims, kernel_size=1, use_bias=False)\n",
    "        self.attention = keras.layers.Attention(causal=self.causal, use_scale=self.use_scale)\n",
    "        self.out_linear = keras.layers.Conv1D(self.dims, kernel_size=1, use_bias=False)\n",
    "        super().build(batch_input_shape)\n",
    "    def _multi_head_linear(self, inputs, linear):\n",
    "        shape = K.concatenate([K.shape(inputs)[:-1], [self.n_heads, -1]])\n",
    "        projected = K.reshape(linear(inputs), shape)\n",
    "        perm = K.permute_dimensions(projected, [0, 2, 1, 3])\n",
    "        return K.reshape(perm, [shape[0] * self.n_heads, shape[1], -1])\n",
    "    def call(self, inputs):\n",
    "        q = inputs[0]\n",
    "        v = inputs[1]\n",
    "        k = inputs[2] if len(inputs) > 2 else v\n",
    "        shape = K.shape(q)\n",
    "        q_proj = self._multi_head_linear(q, self.q_linear)\n",
    "        v_proj = self._multi_head_linear(v, self.v_linear)\n",
    "        k_proj = self._multi_head_linear(k, self.k_linear)\n",
    "        multi_attended = self.attention([q_proj, v_proj, k_proj])\n",
    "        shape_attended = K.shape(multi_attended)\n",
    "        reshaped_attended = K.reshape(multi_attended, [shape[0], self.n_heads, shape_attended[1], shape_attended[2]])\n",
    "        perm = K.permute_dimensions(reshaped_attended, [0, 2, 1, 3])\n",
    "        concat = K.reshape(perm, [shape[0], shape_attended[1], -1])\n",
    "        return self.out_linear(concat)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:Layer multi_head_attention is casting an input tensor from dtype float64 to the layer's dtype of float32, which is new behavior in TensorFlow 2.  The layer has dtype float32 because it's dtype defaults to floatx.\n",
      "\n",
      "If you intended to run this layer in float32, you can safely ignore this warning. If in doubt, this warning is likely only an issue if you are porting a TensorFlow 1.X model to TensorFlow 2.\n",
      "\n",
      "To change all layers to have dtype float64 by default, call `tf.keras.backend.set_floatx('float64')`. To change just this layer, pass dtype='float64' to the layer constructor. If you are the author of this layer, you can disable autocasting by passing autocast=False to the base Layer constructor.\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:Layer multi_head_attention is casting an input tensor from dtype float64 to the layer's dtype of float32, which is new behavior in TensorFlow 2.  The layer has dtype float32 because it's dtype defaults to floatx.\n",
      "\n",
      "If you intended to run this layer in float32, you can safely ignore this warning. If in doubt, this warning is likely only an issue if you are porting a TensorFlow 1.X model to TensorFlow 2.\n",
      "\n",
      "To change all layers to have dtype float64 by default, call `tf.keras.backend.set_floatx('float64')`. To change just this layer, pass dtype='float64' to the layer constructor. If you are the author of this layer, you can disable autocasting by passing autocast=False to the base Layer constructor.\n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "TensorShape([2, 50, 512])"
      ]
     },
     "execution_count": 77,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Q = np.random.rand(2, 50, 512)\n",
    "V = np.random.rand(2, 80, 512)\n",
    "multi_attn = MultiHeadAttention(8)\n",
    "multi_attn([Q, V]).shape"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.7.6"
  },
  "nav_menu": {},
  "toc": {
   "navigate_menu": true,
   "number_sections": true,
   "sideBar": true,
   "threshold": 6,
   "toc_cell": false,
   "toc_section_display": "block",
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}