Advantage Actor Critic (A2C)

Written and documented in PyTorch

This is a repository of the A2C reinforcement learning algorithm in the newest PyTorch (as of 03.06.2019) including also Tensorboard logging. The agent.py file contains a wrapper around the neural network, which can come handy if implementing e.g. curiosity-driven exploration.

Running should be straightforward, all the command line arguments can be found in utils.py. Running

python ./main.py

should launch the training on Pong.

While trying to immerse into deep reinforcement learning, I created this repo to give you a well documented resource for A2C, as in my opinion, most publicly available repositories are either not self-explanatory or just not documented well.

I would like to list the repos I collected lots of help, please not that many of them offer a much wider range of functionality, which was not the goal in my case.

• https://github.com/ikostrikov/pytorch-a2c-ppo-acktr-gail
• https://github.com/ikostrikov/pytorch-a3c
• https://github.com/lnpalmer/A2C