3bca459d2fc85937a84a17657e19e7f0167d75ff,ml/rl/training/ddpg_trainer.py,DDPGTrainer,train,#DDPGTrainer#Any#Any#Any#,114

Before Change

            prev_max=self.max_action_range_tensor_serving,
        )
        rewards = Variable(torch.from_numpy(training_samples.rewards).type(self.dtype))
        next_states = Variable(
            torch.from_numpy(training_samples.next_states).type(self.dtype)
        )
        time_diffs = torch.tensor(training_samples.time_diffs).type(self.dtype)
        discount_tensor = torch.tensor(np.full(len(rewards), self.gamma)).type(
            self.dtype
        )

After Change

        actions = Variable(actions)

        if isinstance(training_samples.next_states, torch.Tensor):
            next_states = training_samples.next_states.type(self.dtype)
        else:
            next_states = torch.from_numpy(training_samples.next_states).type(
                self.dtype
            )
        next_states = Variable(next_states)

        // As far as ddpg is concerned all actions are [-1, 1] due to actor tanh
        actions = rescale_torch_tensor(
            actions,

Instances