83ea0caeb48b0bcf50f1302a7e0f67ba3e7ce48d,ml/rl/prediction/dqn_torch_predictor.py,ParametricDqnTorchPredictor,policy,#ParametricDqnTorchPredictor#Any#Any#,121

Before Change

            1.0 - possible_actions_presence.reshape(1, -1)  // type: ignore
        ) * 1e10

        q_scores_softmax_numpy = (
            masked_softmax(
                q_scores.reshape(1, -1),
                possible_actions_presence.reshape(1, -1),
                self.softmax_temperature,
            )
            .detach()
            .numpy()[0]
        )
        if (
            np.isnan(q_scores_softmax_numpy).any()
            or np.max(q_scores_softmax_numpy) < 1e-3
        ):
            q_scores_softmax_numpy[:] = 1.0 / q_scores_softmax_numpy.shape[0]

        return DqnPolicyActionSet(
            greedy=int(torch.argmax(q_scores)),
            softmax=int(np.random.choice(q_scores.size()[1], p=q_scores_softmax_numpy)),
        )

    def policy_net(self) -> bool:
        return False

After Change

        )
        q_scores = q_scores.reshape(1, -1)

        return self.policy_given_q_values(
            q_scores, self.softmax_temperature, possible_actions_presence
        )

    @staticmethod
    def policy_given_q_values(
        q_scores: torch.Tensor,

Instances