55 commits

Author	SHA1	Message	Date
Sven	8b16847c02	Get utils ready for better Agent torch support. (#6561)	2019-12-30 12:27:32 -08:00
Matthew A. Wright	0110941de5	rllib: use pytorch's fn to see if gpu is available (#5890)	2019-10-12 00:13:00 -07:00
Matthew A. Wright	4aa06918ae	Qmix on gpu and with non-stacked-obs environment state support (#5751)	2019-10-08 13:18:07 -07:00
Matthew A. Wright	3131e1742d	[rllib] Qmix off by 1 in double Q calculation (#5731) ... * Qmix fix. -Current version of double Q learning is incorrect; it selects actions at timestep t instead of t+1 when computing the t+1 Q value. * Allow extra obs dict keys * Move Q-value-computing replay code to own function * Run the autoformatter * use better terms in comments ("policy" network instead of "live" network)	2019-09-18 18:12:30 -07:00
Eric Liang	5d7afe8092	[rllib] Try moving RLlib to top level dir (#5324)	2019-08-05 23:25:49 -07:00