question about " start_use_mix_training_steps " and " mixed_value_threshold "

[sarsigmadelta]

Firstly thanks for sharing such a wonderful proj , i get a confusion in the code: list "top_new_masks" in the make_batch function of batch_worker.py(about line 173) , append data to it with top_new_masks.append(int(sample_idx > collected_transitions - self.mixed_value_threshold)), list "top_new_masks" was used in the agents/base.py ,this_target_values = target_values * top_value_masks.unsqueeze(1).repeat(1, unroll_steps + 1) \ + search_values * (1 - top_value_masks).unsqueeze(1).repeat(1, unroll_steps + 1) why this top_new_masks be helpful ？ and what " start_use_mix_training_steps " and " mixed_value_threshold " does ？ thanks again， and looking forward to your response

[Shengjiewang-Jason]

Hi, man. I understand your question. You can see that $T_1$ is the start_use_mix_training_steps and $T_2$ is mixed_value_threshold in our paper. I guess this will help you.