search

eeyhsong / EEG-Conformer

EEG Transformer 2.0. i. Convolutional Transformer for EEG Decoding. ii. Novel visualization - Class Activation Topography.
GNU General Public License v3.0
389 stars 55 forks source link

About SEED dataset #5

Closed FormerRiptide3 closed 1 year ago

FormerRiptide3 commented 1 year ago

Can you please share your preprocessing and training code about the SEED dataset? Especially in the data loading and augmentation part, many parameters are hard-coded, its hard to understand.

384068026 commented 1 year ago

您好，我的研究方向是脑电信号情绪识别，您提到的SEED系列数据集我先前也进行过大量试验，关于本项目作者的另一个项目EEG-Transformer项目我也有所研究，针对您的问题，我似乎有解决方案，但是想与您沟通一下，如您同意，请添加我的QQ：384068026

eeyhsong commented 1 year ago

Hello, @Riptide3, Sorry, the code for SEED has not been prepared well. This part also uses only band-pass filter and standardization for preprocessing, without data augmentation (due to enough data). And we choose 5-fold cross-validation instead of using 9 trials for training and 6 trials for test.

eeyhsong commented 1 year ago

Hello, @384068026 You can directly use Confomer, which is more reasonable than the previous one.

384068026 commented 1 year ago

非常感谢您的答复，我想问的问题是seed数据集上进行实验只需要将数据按照1s的时间窗大小来切片然后堆叠成getData.m输出的那个结构就可以了是吗？模型内的参数什么的需要调参吗？然后用kflod来分配数据就可以了吧？您方便提供一些这些代码吗？如果您还未打算开源这一部分，我将自己尝试。非常感谢您的贡献

eeyhsong commented 1 year ago

@384068026 您好，是这样的。seed上我没有进行较详细的调参，感兴趣的话您可以尝试一下。这部分没有整理好，暂时不release了，非常抱歉