b9b54d4f30ff65cf1c54dc0cf90c938b48c44f90,reagent/training/qrdqn_trainer.py,QRDQNTrainer,train,#QRDQNTrainer#Any#,144

Before Change

        target_Q = rewards + discount_tensor * not_done_mask * next_qf

        with torch.enable_grad():
            current_qf = self.q_network.dist(state).q_values

            // for reporting only
            all_q_values = current_qf.mean(dim=2).detach()

After Change

        target_Q = rewards + discount_tensor * not_done_mask * next_qf

        with torch.enable_grad():
            current_qf = self.q_network.dist(state)

            // for reporting only
            all_q_values = current_qf.mean(2).detach()

Instances