PARL/examples/Prioritized_DQN

• Prioritized Experience Replay
• Reproduced Results
• How to use
• • Dependencies:
  • Start Training:

Prioritized Experience Replay

Reproducing paper Prioritized Experience Replay.

Prioritized experience replay (PER) develops a framework for prioritizing experience, so as to replay important transitions more frequently. There are two variants of prioritizing the transitions, rank-based and proportional-based. Our implementation is the proportional variant, which has a better performance, as reported in the original paper.

Reproduced Results

Results have been reproduced with Double DQN on following three environments:

How to use

Dependencies:

• paddlepaddle>=1.8.5
• parl
• gym[atari]==0.17.2
• atari-py==0.2.6
• tqdm
• ale_python_interface

Start Training:

Train on BattleZone game:

python train.py --rom ./rom_files/battle_zone.bin

To train on more games, you can install more rom files from here.