Kaggle competition: https://www.kaggle.com/t/c88b47a9221b43dc95f73cd696b5695f

This is a classification task with 7 categories, involving two files:

• train.npz
• test.x.npz

The training file contains 8,000 samples, each consisting of 100 features and 7 labels. The testing file comprises 2,000 samples, each having 100 features.

To load the files, you can use the following code:

import numpy as np
train = np.load('train.npz')
test = np.load('test.x.npz')
train_x, train_y = train['x'], train['y']
test_x = test['x']

Your objective is to classify and predict the 2,000 test data samples and obtain 2,000 rows of labels ranging from 0 to 6.

For example:

pred = model(test_x)  # 2000 x 1 matrix
print(pred)
# [[0],
#  [6],
#  [2],
#  ...,
#  [1],
#  [3]]
np.savetxt(
    "pred.csv",
    np.asarray(list(enumerate(pred.reshape(-1)))),
    fmt="%d",
    delimiter=",",
    header="id,label",
    comments="",
)

You could see a baseline model in src/simp/baseline.py. It is a simple MLP model with 3 hidden layers, dropout and batch normalization. The accuracy is about 0.8.

To run it:

You need to install the following packages:

dependencies = [
    "jax>=0.4.6",
    "jaxlib>=0.4.6",
    "dm-haiku>=0.0.9",
    "optax @ git+https://github.com/deepmind/optax.git",
    "numpy>=1.24.2",
    "rich>=13.3.2",
]

Or using pdm https://pdm.fming.dev/latest/ to install dependencies:

pdm install

You could run the baseline model by:

python src/simp/baseline.py

# or if you use pdm
pdm run python src/simp/baseline.py

We evaluate the predition txt using accuracy. The accuracy is defined as:

def accuracy(pred, label):
    return np.mean(pred == label)

You could see an example in src/simp/test_results.py.