search

Want2Vanish / FBMSNet

This is the PyTorch implementation of the FBMSNet architecture for EEG-MI classification.
30 stars 7 forks source link

The accuracy reproduced by source code is lower than accuracy presented in the paper. #3

Open realblack0 opened 1 year ago

realblack0 commented 1 year ago

Hello,

I have been working on reproducing the experiments using the provided source codes. To solve an error in saveData.py, I made some minor changes to generate the events object as FBCSP code. Additionally, I followed the given random seed for consistency.

However, I am encountering a challenge as the resulting average accuracies are different from those reported in the paper. Specifically, the achieved average accuracies were 77.62% and 69.69% for the BCIC IV 2a and OpenBMI datasets, respectively. These values are lower than the accuracies documented in the paper.

I am puzzled by the discrepancy, especially since I have ensured that the random seed remains unchanged. I am reaching out to kindly request guidance on how to reproduce the same results as presented in the paper.

Any insights or assistance you can provide would be greatly appreciated. Thank you for your time and support.

Want2Vanish commented 1 year ago

Hi, thanks for your interest in our work.

As for your puzzle, first of all I would like to know if you have successfully reproduced the experiment results from FBCNet? Related link：https://github.com/ravikiran-mane/FBCNet

In addition, it is very important to make sure that the source code runs in the same environment as specified in the environment.txt file.

realblack0 commented 1 year ago

Hi, thanks for your interest in our work.

As for your puzzle, first of all I would like to know if you have successfully reproduced the experiment results from FBCNet? Related link：https://github.com/ravikiran-mane/FBCNet

In addition, it is very important to make sure that the source code runs in the same environment as specified in the environment.txt file.

Thank you for the immediate response. I have not considered the environment.txt. As you said, I will set the same environment and check the related link.

Want2Vanish commented 1 year ago

Hi, thanks for your interest in our work. As for your puzzle, first of all I would like to know if you have successfully reproduced the experiment results from FBCNet? Related link：https://github.com/ravikiran-mane/FBCNet In addition, it is very important to make sure that the source code runs in the same environment as specified in the environment.txt file.

Thank you for the immediate response. I have not considered the environment.txt. As you said, I will set the same environment and check the related link.

Creating a source code runtime environment using the "environment.txt" file may have some problems, I'll be updating that document in a couple days and uploading a new document about how to create a source code runtime environment. If you have any other questions, please let me know.

Want2Vanish commented 1 year ago

I've updated the env.txt file and added the runtime environment creation steps required to run the source code to the readme.

f348564110 commented 4 months ago

Hi, thanks for your interest in our work. As for your puzzle, first of all I would like to know if you have successfully reproduced the experiment results from FBCNet? Related link：https://github.com/ravikiran-mane/FBCNet In addition, it is very important to make sure that the source code runs in the same environment as specified in the environment.txt file.

Thank you for the immediate response. I have not considered the environment.txt. As you said, I will set the same environment and check the related link.

Hello, I have already set up the environment provided by the author and replicated the results of FBCNet. However, while reproducing the results for BCI IV 2a in this paper, the accuracy remains at only 77.89%. May I kindly ask if you were able to successfully replicate it, and if so, what adjustments did you make?