2c7a2997c28d0be732e4157e924dce87bcf56236,ml/rl/training/discrete_action_trainer.py,DiscreteActionTrainer,get_q_values_all_actions,#DiscreteActionTrainer#Any#Any#,192

Before Change

            trainer"s TargetNetwork to compute Q values.
        
        if use_target_network:
            return self.target_network.target_values(states)
        else:
            all_q_values = C2.NextBlob("all_q_values")
            MakeForwardPassOps(
                C2.model(),

After Change

        all_q_values = C2.NextBlob("all_q_values")
        if use_target_network:
            if self.conv_target_network is not None:
                conv_output_blob = C2.NextBlob("conv_output")
                self.conv_target_network.make_conv_pass_ops(
                    C2.model(),
                    states,
                    conv_output_blob,
                )
                states = conv_output_blob
            self.target_network.make_forward_pass_ops(
                C2.model(),
                states,
                all_q_values,
                True,
            )
        else:
            if self.conv_ml_trainer is not None:
                conv_output_blob = C2.NextBlob("conv_output")
                self.conv_ml_trainer.make_conv_pass_ops(
                    C2.model(),
                    states,
                    conv_output_blob,
                )

Instances