DQN

Reproduce DQN with PARL

Based on PARL, the DQN algorithm of deep reinforcement learning has been reproduced, reaching the same level of indicators as the paper in Atari benchmarks.

• DQN in Human-level Control Through Deep Reinforcement Learning

Atari games introduction

Please see here to know more about Atari games.

Benchmark result

How to use

Dependencies:

• python2.7 or python3.5+
• paddlepaddle>=1.0.0
• parl
• gym
• tqdm
• atari-py
• ale_python_interface

Start Training:

# To train a model for Pong game
python train.py --rom ./rom_files/pong.bin

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