khangt1k25 / MADE

Implement MADE

An implementation for MADE: Masked Autoencoder for distribution estimation

1. Data: Download and preprocessing

• MNIST: https://github.com/mgermain/MADE/releases/download/ICML2015/binarized_mnist.npz
• FOUR SHAPES https://www.kaggle.com/smeschke/four-shapes -> crop to (100,100) and turn to numpy file data.npy
• EMNIST: https://www.nist.gov/itl/products-and-services/emnist-dataset -> emnist-letters.mat

2. Implement: Implement in MADE_pytorch.ipynb. All libs in this implementation are in the top line of codes.
3. Ref: https://arxiv.org/abs/1502.03509 https://github.com/karpathy/pytorch-made http://bjlkeng.github.io/posts/autoregressive-autoencoders/ ... (See more at the end of the report)