mabhijithn / irregulars-neureka-codebase Goto Github PK

I encounter an error while running the IC label filtering procedure on the 03_tcp_ar_a configuration in the training set. In particular, the error occurs while processing the subject 00003282 (no other subject causes an error). The following error is received.

Any guideline as to how to resolve this issue?

After reading your article about the Neureka Challenge 2020, I would like to ask some questions about the code of seizure detection algorithm.

1. Prediction results of Unet. After I ran the file (irregulars-neureka-codebase/evaluate/2-unet-prediction.py), I found that the length of the ground truth and the predicted result were not the same. I want to know what caused the issue.

2. Run file order. After I have trained Unet (/train/3-train-unet), should I execute unet prediction (/evaluate/2-unet-prediction) next to get prediction results ? And then I use this result to train the LSTM, right?

3. The difference between the function - build_unet() and build_windowfree_unet(). I found that you used different Unet functions in the training and evaluation. Why do you use Window free? Can I use the function build_unet() to make a prediction?

Best，looking forward to your reply，

Congrats on achieving a great score!! The presentation was really good and gave us a lot of insights. Thanks for that! :D

In Line no:

The function used here is os.joind. However, there are no functions like joind in the os module. Should it be os.path.join?

Also, the open issue : #2, where input and output doesn't exactly match?

I have a question about the code in the file in train/4-train-lstm.py

When you define the function, the function just return x, which refers to the three pre-computed prediction. However, in the function train(), you use the function prepare_file() to return x & y. What does y stand for? I guess y stands for reference file, right?

Thanks for your great work and contribution to the EEG analysis society.

I am interested to know the specs of the system used to train the model. Clock Speed, RAM, HDD disk space, GPU (in case it was used), etc..

Also, what was the average time to train the model on the whole train dataset?

Thank you

Thank you for your contribution to EEG.
I'm new to this,when i run your code ,i have a lot of version issues,so i want to know all the versions that you list in the python_requirement.txt.
I also want to know how to download the dataset

I am looking forward to your early reply,thanks.

The code for generating segments seems to be missing a window.

I believe line 49 should be set to

n = 1 + (array.shape[0] - self.window_size)//self.stride

Hi,

I have tried to train the U-Net on the data view resulting from the subspace projection filtering (Section II-A1 in the paper).

I have first used the code in irregulars-neureka-codebase/training/3-DNN/3-load-data.py with the following changes (both variables set to True)

includes_tse = True # Boolean flag, whether .tse files with ground-truth labels can also be loaded
wiener = True # Boolean flag, whether Wiener filtering is applied

and used the wiener filters provided in irregulars-neureka-codebase/library/filters.pickle to filter the data.

Then, I used irregulars-neureka-codebase/training/3-DNN/3-train-unet.py to train the unet using the data generated from 3-load-data.py.

The code runs with no errors, however the loss is equal to NaN.

Have you also experianced this issue?

Many thanks in advance.