chinnadhurai/pytorch

myTorch

A pyTorch research framework for Memory Augmented Neural Networks and Reinforcement Learning.

Major directories

• myTorch/utils - contains basic experiment manager, tensorboard logger, and util functions.
• myTorch/memory - contains memory cells like LSTM, GRU.
• myTorch/rllib - contains standard implementations of RL agents.
• myTorch/environment - contains several environments that can be linked to the RL agents.
• myTorch/task - contains data iterators for tasks used in the projects.
• myTorch/projects - contains projects based on myTorch.

Setup

• run pip install -r requirements.txt
• set an environment variable called 'MYTORCH_SAVEDIR' to a folder where you want the output logs and model checkpoints to be saved.
• set an environment variable called 'MYTORCH_DATA' to a folder where you want to keep all your data.