search

cathywu / rllab

rllab is a framework for developing and evaluating reinforcement learning algorithms, fully compatible with OpenAI Gym.
Other
1 stars 0 forks source link

Multiagent point env #8

Open cathywu opened 7 years ago

cathywu commented 7 years ago

This is an issue to track the comparison between the baselines for MultiagentPointEnv.

Without parameter tuning and for MultiagentPointEnv(d=1, k=6) 2017-04-21-multiagentpointenv 1 6 -baselinecomparison

ActionDependentGaussianMLPBaseline appears to converge about 2x faster than GaussianMLPBaseline. Note: no tuning has been done whatsoever at this point. This is 26 runs so far across 3 step sizes (random seed 1 (so far)).

Running with commit 4e11e7ff29e.