This project is the implementation for Natural and Dueling DQN for Breakout.
pip install -r requirements.freezed
./run_dueling.shYou can modify the parameters according to your own needs and refer to config.py for details.
We have trained Dueling DQN for 50m steps and we evaluated the reward value every 0.1m steps. We obtained the best model with reward 394.7 at step 397. Refer to models/DuelingDQN/rewards.txt for details.
On the basis of https://gitee.com/goluke/dqn-breakout#dqn-for-breakout, we mainly modify main.py, utils_drl.py, utils_model.py, run.sh and add config.py, run_dueling.sh, run_pretrain.sh.