chainer-ppo

Reproduction codes of Proximal Policy Optimization (PPO) with chainer

About

This repo is a PPO reproduction codes writen with chainer. See this original paper for details

Training the network

Choose the params and run below command. The default parameters are set for running in atari environment.

Example:

python3 main.py --env-type='atari' 

For the detail of the parameters check the code or type

python3 main.py --help

Results

Atari

Breakout

Small model (2 conv layers model)

$ python3 main.py --env-type='atari' --test-run --model-params=trained_results/atari/breakout/small/final_model --atari-model-size='small'

result	score

Large model (3 conv layers model)

python3 main.py --env-type='atari' --test-run --model-params=trained_results/atari/breakout/large/final_model --atari-model-size='large'

result	score

Zaxxon

Large model (3 conv layers model)

python3 main.py --env-type='atari' --test-run --model-params=trained_results/atari/zaxxon/large/final_model --atari-model-size='large' --env='ZaxxonNoFrameskip-v4'

result	score

Space Invaders

Large model (3 conv layers model)

python3 main.py --env-type='atari' --test-run --model-params=trained_results/atari/space_invaders/large/final_model --atari-model-size='large' --env='SpaceInvadersNoFrameskip-v4'

result	score

Mujoco

Sorry in progress...