I implemented an algorithm called Ephemeral Value Adjustments (EVA) using pfrl.
- python >= 3.8
- pfrl
- OpenAI gym
- atari-py >= 0.2.6
python train.py [options]
--env
: Environment name of the Atari.- Eg :
--env PongNoFrameskip-v4
- Eg :
--gpu
: If 0, use the GPU. If the number is less than 0, the GPU is not used.--agent
: Choose DQN and EVA.
- Reduce GPU memory usage
- More experiments.