7d2a67412daa0abb3ce2339a8f716a4ca51791e4,src/garage/torch/algos/vpg.py,VPG,_compute_kl_constraint,#VPG#Any#,366

Before Change

        with torch.no_grad():
            old_dist = self._old_policy(obs)

        new_dist = self.policy(obs)

        kl_constraint = torch.distributions.kl.kl_divergence(
            old_dist, new_dist)

After Change

        with torch.no_grad():
            old_dist = self._old_policy(obs)[0]

        new_dist = self.policy(obs)[0]

        kl_constraint = torch.distributions.kl.kl_divergence(
            old_dist, new_dist)

Instances