policy-gradient-methods

policy-gradient-methods copied to clipboard

Using sampled entropy rather than analytic entropy.

zafarali

zafarali

According to fig 3 here https://arxiv.org/abs/1707.06347 VPG should be able to do ok after about 0.5M steps

zafarali

Seems like discrete spaces with discrete actions is not doing well at all. Might be a wrapper problem

zafarali