yangyi0318 / ptan

PyTorch Agent Net: reinforcement learning toolkit for pytorch

PTAN

PTAN stands for PyTorch AgentNet -- reimplementation of AgentNet library for PyTorch

This library was used in "Deep Reinforcement Learning Hands-On" book, here you can find sample sources.

Code branches

The repository is maintained to keep dependency versions up-to-date. This requires efforts and time to test all the examples on new versions, so, be patient.

The logic is following: there are several branches of the code, corresponding to major pytorch version code was tested. Due to incompatibilities in pytorch and other components, code in the printed book might differ from the code in the repo.

At the moment, there are the following branches available:

• master: contains the code with the latest pytorch which was tested. At the moment, it is pytorch 1.7.
• torch-1.3-book-ed2: code printed in the book (second edition) with minor bug fixes. Uses pytorch=1.3 which is available only on conda repos.
• torch-1.7: pytorch 1.7. Merged with master.

All the branches uses python 3.7, more recent versions weren't tested.

Installation

From sources:

python setup.py install

From pypi:

pip install ptan

From github:

pip install pip install git+https://github.com/Shmuma/ptan.git 

Requirements

• PyTorch: version 1.1.0 is required
• PyTorch Ignite: provides extra bindings for ignite
• OpenAI Gym: pip install gym gym[atari]
• Python OpenCV: pip install opencv-python
• TensorBoardX: pip install tensorboardX

Note for Anaconda Python users

To run some of the samples, you will need these modules:

conda install pytorch torchvision -c pytorch
pip install tensorboard-pytorch
pip install gym
pip install gym[atari]
pip install opencv-python

Documentation

• Ptan introduction

Random pieces of information

• ExperienceSource vs ExperienceSourceFirstLast: Shmuma#17 (comment)