Closed XCZchaos closed 3 months ago
Hello, @XCZchaos, Sorry for the late reply. In my view, the preprocessing like band-pass is very important, and different participants are variant to different frequency bands, which may introduce the results.
Thanks for your answer, but I used the getdata file on another of your projects for filtering and found that the effect is still not as good as the effect without filtering
Hello, @XCZchaos, 你好 Sorry for the late reply. In my view, the preprocessing like band-pass is very important, and different participants are variant to different frequency bands, which may introduce the results.对不起,回复晚了。在我看来,像带通这样的预处理非常重要,不同的参与者是不同频段的变体,这可能会引入结果。
Sorry. Do you mean the results without band-pass filter get the best results?
hello @XCZchaos Have you added the data augmentation part?
hello @XCZchaos Have you added the data augmentation part?您好,您是否添加了数据增强部分?
Sorry for the late reply, thank you for your multiple replies. After many experiments, it is found that the effect of filtering is better, and I use data augmentation, but I suggest that you add data augmentation in another version of the code, I think it will make careless people like me forget to add data augmentation and lead to poor model performance
Sorry. Do you mean the results without band-pass filter get the best results?不好意思。您的意思是没有带通滤波器的结果会得到最好的结果吗?
I think there may be something wrong with my python code. I used your matlab code and found that the effect was consistent with the original paper
It's good to see the final results well. Thanks a lot for your suggestion! We will try to make the code clearer.
Best wishes. 🤝
Dear author, I recently found a problem of low accuracy in your model. A01acc was 0.78, A02acc was 0.50, and A03 was also about 0.86, but the acc of S5 was unexpectedly up to 75. I used mne for pre-processing and extracted 2-6s and motion imagination data with ID 768. In addition, data standardization was carried out in the pre-processing stage (the get data module in the original paper has been adjusted to the data after its own pre-processing), but I did not modify other parts of the code. avg acc is around 73 Another occurrence is that not doing bandpass filtering seems to improve the accuracy (multiple experimental results). Here is my python code (unfiltered)