783e415c29c4a4124ea426c824864a32b0f8ee71,src/garage/tf/policies/gaussian_mlp_policy.py,GaussianMLPPolicy,get_action,#GaussianMLPPolicy#Any#,210

Before Change

    @overrides
    def get_action(self, observation):
        flat_obs = self.observation_space.flatten(observation)
        mean, log_std = [x[0] for x in self._f_dist([flat_obs])]
        rnd = np.random.normal(size=mean.shape)
        action = rnd * np.exp(log_std) + mean
        return action, dict(mean=mean, log_std=log_std)

After Change

        mean, log_std = self._f_dist([flat_obs])
        rnd = np.random.normal(size=mean.shape)
        sample = rnd * np.exp(log_std) + mean
        sample = self.action_space.unflatten(sample[0])
        mean = self.action_space.unflatten(mean[0])
        log_std = self.action_space.unflatten(log_std[0])
        return sample, dict(mean=mean, log_std=log_std)

Instances