e91a9941d0ee6d1198e4b05bbd39ce16b5c0d569,ml/rl/training/discrete_action_trainer.py,DiscreteActionTrainer,get_max_q_values,#DiscreteActionTrainer#Any#Any#Any#,137

Before Change

            next_states, use_target_network
        )

        if possible_next_actions is not None:
            mask = np.multiply(
                np.logical_not(possible_next_actions),
                self.ACTION_NOT_POSSIBLE_VAL
            )
            q_values += mask

        return np.max(q_values, axis=1, keepdims=True)

    def get_q_values(
        self, states: np.ndarray, actions: np.ndarray

After Change

            self.ACTION_NOT_POSSIBLE_VAL,
            broadcast=1,
        )
        q_values = C2.Add(q_values, inverse_pna)

        q_values_max = C2.ReduceBackMax(
            q_values,
            num_reduce_dims=1,
        )
        return C2.ExpandDims(q_values_max, dims=[1])

    def get_q_values_all_actions(
        self,
        states: str,

Instances