search

UlysseCoteAllard / MyoArmbandDataset

GNU General Public License v3.0
86 stars 40 forks source link

PyTorchImplementation #10

Open GuiyinLi opened 2 years ago

GuiyinLi commented 2 years ago

Hello, I have some questions.

  1. The concrete version of PyTorch and Third-party Python libraries
  2. I just found the "CWT" and "RawEnhancedConvNet" in "PyTorchImplementation". Didn't you implement this Spectrogram in the PyTorch because of porting from Theano to PyTorch
  3. Is the PELU empolyed in PyTorch networks? Because I see the "PELU.py" in the dir.
GuiyinLi commented 2 years ago

In addition, I think that you can create "Theano" then move the "CWT" and "Spectrogram" of the root directory to the “Theano” dir, which might make your project become more clear.

UlysseCoteAllard commented 2 years ago

Hi, thanks for reaching out.

  1. I have to admit, back then I forgot to record this information and so I do not remember right now. I will go back in the future and find out the earliest version for each library that can be used to run the code (or update it so that it run with the current versions (and note them)). I would do that for the PyTorch version, not for Theano. However, this won't be in the next few weeks as I am in a writing rush period.
  2. The Spectrogram was not implemented in PyTorch, but I could do it when I attend to point number 1 as it would be relatively easy.
  3. In the PyTorch implementation, I am using Parametric RELU if I remember correctly because the implementation I did of PELU was too slow. I might check if there is an efficient implementation for PELU in PyTorch now.
  4. I agree, back then Theano was still relevant, but this is not the case anymore. I'll most likely do a refactoring when I have time and to address UlysseCoteAllard/MyoArmbandDataset#9

Again, thank you for reaching out and I'll try to get to it as soon as possible, however, I cannot give you a more precise timeline at the moment, unfortunately.

GuiyinLi commented 2 years ago

It doesn't matter. I'm trying to run the PyTorchImplementation with my current PyTorch of the new version and py38. If the run is successful, I also will try to implement the Spectrogram of PyTorch although I do not understand its theory. Of course, I'm looking forward to your perfection. Good Luck!

53LostNFound commented 3 weeks ago

It doesn't matter. I'm trying to run the PyTorchImplementation with my current PyTorch of the new version and py38. If the run is successful, I also will try to implement the Spectrogram of PyTorch although I do not understand its theory. Of course, I'm looking forward to your perfection. Good Luck! Have you fully implemented the files in this project? I am still confused about some issues, such as I can't find datasets_processed/saved_dataset_training.npy