Neural networks explained with moving pictures and Mandarin

What it does

This repository is a structured collection of PyTorch tutorials that starts with tensor basics and ends with reinforcement learning and generative adversarial networks. Each lesson is a standalone Python script—plus contributed Jupyter notebooks—covering topics like regression, CNNs, RNNs, and AutoEncoders. The README embeds animated GIFs of training results so you can see what success looks like before running anything.

The interesting bit

The project doubles as the code companion to 莫烦Python, a Chinese-language learning platform, so every method comes with video and text explanations in Mandarin. That makes it a rare example of a major ML tutorial series built natively for Chinese speakers rather than simply translated.

Key highlights

• Covers a wide arc: from torch and numpy basics up through DQN, A3C, and conditional GANs
• README includes result GIFs for regression, classification, CNN, RNN, AutoEncoder, GAN, Dropout, and Batch Normalization
• Provides both raw .py tutorial files and a separate notebook conversion (courtesy of liufuyang)
• Explicitly targets Chinese speakers with linked video and text tutorials on mofanpy.com
• Also points to a matching TensorFlow tutorial for framework comparison

Caveats

• Some material shows its age: the curriculum still features Variable, a PyTorch class since merged into standard tensors
• The “Others” section—covering dynamic graphs, GPU training, Dropout, and Batch Normalization—is marked WIP
• The A3C lesson lives in a separate repository rather than inside this one

Verdict

Worth bookmarking if you are a PyTorch beginner, especially one who reads Chinese or learns best by watching animated loss curves converge. Skip it if you are already comfortable building CNNs and GANs and need advanced engineering patterns.