Model Zoo for Deep Reinforcement Learning. Place for trained deep RL models, with focus on DQN and ALE.
Currently looking for trained models. If you have some consider contributing.
- https://github.com/tambetm/simple_dqn - DQN ALE implementation, using Neon. Has trained models for Breakout and Pong.
The following are in the format gamename_datetrained_epoch_gitsha1. The sha1's mentioned are from https://github.com/alito/deep_q_rl or from https://github.com/alito/deep_q_rl/tree/rl_glue for the NIPS models. You may be able to watch them play using newer code as long as you select the right model (Nature unless otherwise specified).
- http://organicrobot.com/deepqrl/atlantis_20151122_e78_94f7d724f69291ecfe23ac656aef38c63a776b8d.pkl.gz
- http://organicrobot.com/deepqrl/boxing_20150731_e41_c9e7447b005e23202055a22c5b902cf244203ebe.pkl.gz
- http://organicrobot.com/deepqrl/crazy_climber_20151018_e90_54374b2c86698bd8c71ca8d5936404340c0bea2d.pkl.gz
- http://organicrobot.com/deepqrl/enduro_20150211_e70_nips_7ca3c49a076654d369d97569f2fe2847509503e8_maybe.pkl.gz
- http://organicrobot.com/deepqrl/qbert_20150213_e185_nips_7ca3c49a076654d369d97569f2fe2847509503e8_maybe.pkl.gz
- http://organicrobot.com/deepqrl/river_raid_20150226_e157_nips_48afab9636183c8064f6087590a18990e66dd05c_maybe.pkl.gz
- http://organicrobot.com/deepqrl/seaquest_20150816_e127_b56bfb252320302f498a5b88ad95b55ad1b73e53.pkl.gz
- http://organicrobot.com/deepqrl/space_invaders_20151126_e161_doubleq_94f7d724f69291ecfe23ac656aef38c63a776b8d.pkl.gz (trained using doubleq)
Depending on the size of your models you can either submit pull request with files/links or contact me directly via email. Please provide details about the training procedure (architecture, training parameters, link to the implementation).
For now there there is no specific format for those details; I will prepare one after several models are submitted.