9 Commits
Author SHA1 Message Date
Mike Clark ed943a6356 Use logits in CategoricalActorCriticNet
A really minor change. Should we input logits instead of probs into `Categorical`? It will do less converting from logits and back, leading to increased stability during training (theoretically).
2018-06-15 08:24:14 +08:00
Shangtong Zhang 20f34935e5 Update pixel ddpg params 2018-05-18 15:29:17 -06:00
Shangtong Zhang 93a5c3fc04 Minor update 2018-05-18 14:24:45 -06:00
Shangtong Zhang f6d0c4d260 Remove unused wrapper 2018-05-17 23:37:50 -06:00
Shangtong Zhang 84f7911bb4 Upgrade all actor critic methods 2018-05-17 23:37:17 -06:00
Shangtong Zhang 08038f9533 Upgrade DDPG 2018-05-17 17:17:12 -06:00
Shangtong Zhang 06876d1269 Add Option-Critic 2018-05-07 22:01:21 -06:00
Shangtong Zhang fa294cd532 Fix import 2018-05-07 21:24:52 -06:00
Shangtong Zhang fb95820803 Package the project 2018-05-07 16:49:01 -06:00