Machine learning experiments and implementations

There are a couple of implementations of GANs using PyTorch. They failed to generate decent looking images for any combination of hyperparameters, except in the case of MNIST and fashion MNIST. This is probably due to some combination of the reduced size and complexity of the images. To improve on these models, the best move would probably be to use an architecture very similar to those used in the literature.

The comma_speedchallenge directory contains code to solve Comma.ai's speed challenge to infer vehicle speeds from dash camera video frames.

The NLP directory contains natural language processing experiments, as well as the solutions to Stanford's CS 224n Natural Language Processing with Deep Learning course.

There is also a file using deep reinforcement learning for the inverted pendulum example.