Deep Deterministic Policy Gradient on PyTorch

Overview

The is the implementation of Deep Deterministic Policy Gradient (DDPG) using PyTorch. Part of the utilities functions such as replay buffer and random process are from keras-rl repo. Contributes are very welcome.

Dependencies

• Python 3.4
• PyTorch 0.1.9
• OpenAI Gym

Run

• Training : results of two environment and their training curves:

  • Pendulum-v0

  $ ./main.py --debug

  

  • MountainCarContinuous-v0

  $ ./main.py --env MountainCarContinuous-v0 --validate_episodes 100 --max_episode_length 2500 --ou_sigma 0.5 --debug

  

• Testing :

$ ./main.py --mode test --debug

TODO

Add batch normalization