Policy-Gradient-and-Actor-Critic-Keras | Simple implementation of Policy Gradient | Reinforcement Learning library

Training an agent to play Pong. To train an agent playing Pong with PG, simply run. You can train the agent to play games different from Pong by using argument --env_name [Atari Game Env Name]. But you should modify some part of all codes in order to fit the given environment. To modify any parameters of model/training progress, please modify agent_pg.py. Test the agent's performance on Pong. By running the following command, you can get agent's average score in 30 episode. Testing can be performed with the pretrained model training by default or with the model you trained by adding argument --test_pg_model_path [your model path]. To visualize the gaming progress, add --do_render to the end. You can also save it to vedio with --video_dir [path to save] (set smaller testing episode before doing so). Playing Cartpole or other Atari games. Agents playing Cartpole with Policy Gradient or Actor-Critic is also in agent_dir/, run (modify) them in order to play Cartpole (other games). Testing is not supported, but can be done easily via implementing functions declared in agent.py.
Training an agent to play Pong To train an agent playing Pong with PG, simply run python3 main.py --train_pg You can train the agent to play games different from Pong by using argument --env_name [Atari Game Env Name] But you should modify some part of all codes in order to fit the given environment. To modify any parameters of model/training progress, please modify agent_pg.py.
Test the agent's performance on Pong By running the following command, you can get agent's average score in 30 episode python3 test.py --test_pg Testing can be performed with the pretrained model training by default or with the model you trained by adding argument --test_pg_model_path [your model path] To visualize the gaming progress, add --do_render to the end. You can also save it to vedio with --video_dir [path to save] (set smaller testing episode before doing so)
Playing Cartpole or other Atari games Agents playing Cartpole with Policy Gradient or Actor-Critic is also in agent_dir/, run (modify) them in order to play Cartpole (other games). Testing is not supported, but can be done easily via implementing functions declared in agent.py.