ML notebooks

Deep Learning

• All the notebooks use PyTorch to as the DL framework.
• Tips for training DL models

Computer Vision

• conv_example.py: Python code to train a Convolution NN on custom dataset, using pytorch dataloader. (End2end example)

• GAN.ipynb: Train a DCGAN on CelebA dataset to generate fake faces.

• DCGAN.ipynb: Train a DCGAN on MNIST dataset to generate fake digits.

GAN

• Training challenges: -- Use small batch size especially if model is small, large batch size may train Discriminator very fast during early phase of training.

NLP

• dataprep_bpe.ipynb: Example notebook to load custom Language dataset for Machine translation and tokenize it using Transformers Library with Byte Pair Encoding.

• transformers.ipynb: Train a custom transformer model for Machine Translation

RL

• reinforce_cartpole.ipynb: REINFORCE algorithm from scratch to solve a simple cartpole envionment. Notebook taken from Yandex RL course