Where are saved_dataset_training.npy files?

[karaposu]

I tried to run your code but it is really hard to understand the run order of files. Please update your readme file in a way which describes how to achieve same results as in paper. I think it will be helpful if you can create a notebook.

[amirarjmand93]

You should try to load the dataset at the first stage. Saved dataset file will be created and be ready for inference

[UlysseCoteAllard]

Hi,

Thank you for your comment, I'll update the readme in the following days so that it's clearer which files to run in what order.

In the meantime, amirarjmand93 answer is correct (thank you for answering).

Note however that the results in the paper were obtained using the Theano framework (the original code uploaded in this Github. While a PyTorch implementation is also available (written later), the actual performance might be slightly different due to some differences in implementation between the two frameworks (the PyTorch implementation also does not use the PELU activation function as the custom implementation I had at the time was too slow). In any case, the PyTorch implementation gives you the training loop necessary to run the transfer learning algorithm so that you can easily use your own model.

Finally, I should also point out that an updated version of the TL algorithm is presented in "A Transferable Adaptive Domain Adversarial Neural Network for Virtual Reality Augmented EMG-Based Gesture Recognition" which in our test was more robust in training (the basic idea is to use adversarial domain adaptation in conjunction with the previously described algorithm to make it harder for the network to learn user-specific features).