592c161032bfd697ef6f2f334c5cd007062bec4b,rllib/models/torch/recurrent_net.py,LSTMWrapper,forward,#LSTMWrapper#Any#Any#Any#,156

Before Change

        // Concat. prev-action/reward if required.
        if self.model_config["lstm_use_prev_action_reward"]:
            wrapped_out = torch.cat(
                [
                    wrapped_out,
                    torch.reshape(input_dict[SampleBatch.PREV_ACTIONS].float(),
                                  [-1, self.action_dim]),
                    torch.reshape(input_dict[SampleBatch.PREV_REWARDS].float(),
                                  [-1, 1]),
                ],
                dim=1)

        // Then through our LSTM.

After Change

        prev_a_r = []
        if self.model_config["lstm_use_prev_action"]:
            if isinstance(self.action_space, (Discrete, MultiDiscrete)):
                prev_a = one_hot(input_dict[SampleBatch.PREV_ACTIONS].float(),
                                 self.action_space)
            else:
                prev_a = input_dict[SampleBatch.PREV_ACTIONS].float()
            prev_a_r.append(torch.reshape(prev_a, [-1, self.action_dim]))
        if self.model_config["lstm_use_prev_reward"]:

Instances