e09e6dc678f9b9c198ed361eea0dffa7749bc553,src/garage/torch/algos/vpg.py,VPG,train_once,#VPG#Any#Any#,130

Before Change

                     if self._minibatch_size else len(rewards_flat))
        for epoch in range(self._max_optimization_epochs):
            shuffled_ids = np.random.permutation(len(rewards_flat))
            for start in range(0, len(rewards_flat), step_size):
                ids = shuffled_ids[start:start + step_size]
                loss = self._train(obs_flat[ids], actions_flat[ids],
                                   rewards_flat[ids], advantages_flat[ids])
            logger.log("Mini epoch: {} | Loss: {}".format(epoch, loss))

        self._value_function.fit(paths)

After Change

            tabular.record("/Entropy", policy_entropy.mean().item())

        with tabular.prefix(self._value_function.name):
            tabular.record("/LossBefore", vf_loss_before.item())
            tabular.record("/LossAfter", vf_loss_after.item())
            tabular.record("/dLoss",
                           vf_loss_before.item() - vf_loss_after.item())

Instances