/DQN

Primary LanguagePython

DQN

This is DQN implemented with tensorflow 2. DQN implemented in two version. The original DQN (Nature 2015) and multi-step DQN. The experiment has done under the OpenAI gym cartpole-v1 environment.

Requirements

  • Python 3.8.2
  • tensorflow 2.2.0rc2
  • matplotlib 3.2.1

Usage

Training only original DQN.


> python script.py orgDQN

Training original DQN and multi-step DQN in a sequence.


> python script.py orgDQN multistep

Result

If the average step count exceed 475, the training ends early.


model end episode final average step
Original DQN 488 475.39
Multi-step DQN (n=3) 303 476.4