MLOPTPSU/FedTorch

A question about drawing the plots

Zhg9300 opened this issue · 2 comments

Zhg9300 commented

Hi,
Recently I've read your paper "Distributionally Robust Federated Averaging" carefully and I'm trying to reproduce the experiments in it.
However, I don't know how to get the plots from the codes, I think modifying main.py and run_mpi.py would help but it seems really complicated. Could you please give me some instructions?
Thanks!

mmkamani7 commented

Hi,

Thanks for your interest. The logs can be used to plot the results. There are some tools for this under the tools folder that can parse the log files and plot them like this one https://github.com/MLOPTPSU/FedTorch/blob/main/fedtorch/tools/load_console_records.py
In addition, I am working on adding Tensorboard plots, so there might not be a need for manually plotting the results.

Zhg9300 commented

Hi,

Thanks for your interest. The logs can be used to plot the results. There are some tools for this under the tools folder that can parse the log files and plot them like this one https://github.com/MLOPTPSU/FedTorch/blob/main/fedtorch/tools/load_console_records.py In addition, I am working on adding Tensorboard plots, so there might not be a need for manually plotting the results.

Thanks for your answer! One more question that I'm confused with is that in this paper we're interested in the worst distribution accuracy among all the devices, but it seems that no functions in the codes can be utilized to obtain these accuracy. Could you please give me some instructions?