Chain123
can not run well
Dear authors, thank you for sharing the source code，it's a great job. I have some questions regarding running the code followed the readme guide.
Firstly, is Central.py the baseline? Is it related to the central server model? I had to change the position of model.train() to make it run properly.
Secondly, in the set_train_test_file() of meta-har.py, I changed [self.user_id = self.training_op["train_file"][0].split(os.sep)[-1].split("_")[0] + "act"] to [self.user_id = self.training_op["train_file"][0].split(os.sep)[-1].split("")[0]] because the format in feature_extraction.py for processing the collected dataset does not include '_act'. After making this change, I was able to run meta-har.py successfully. However, I found that with the default parameters, the results were very poor, with accuracies in meta-train and meta-har even below 10%. Additionally, during the running process, I received a warning:
UserWarning:
Detected call of lr_scheduler.step() before optimizer.step().
In PyTorch 1.1.0 and later, you should call them in the opposite
order: optimizer.step() before lr_scheduler.step(). Failure to do this will result
in PyTorch skipping the first value of the learning rate schedule.
See more details at https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate
warnings.warn("Detected call of lr_scheduler.step() before `optimizer.step()".
I have not yet resolved this issue. I am unsure if this warning is causing the poor results or if there is something else I have overlooked.
Thirdly, I have not fully understood the initialization of the server model. Does the server model undergo a training process, or does the local client only pull parameters from the server model for training and updating the server model's initialization? This is because I only see the operation server_model = reptile_meta(har_model.norm_embed, lr, device, utils.pairwiseloss(), 100, 7). Is this the preliminary initialization?
I would greatly appreciate your response.
The above figures are the meta-test test results and meta-train test result during the running process respectively