improve temporal prediction

[Xand04]

Hello! Thank you very much for this very interesting project. I was wondering if it was possible to improve temporal precision of the prediction by decreasing the WIN_SIZE_SEC. However, I get a layer size error (for WIN_SIZE_SEC = 0.975/2).

This is of course depending on the size of the layers of your training model. I was wondering if you have other models available, or an idea to help me work around this issue.

Thanks and have a nice day!

[robertanto]

Have you updated the parameters in the params.py file?

The network adopts a Global Average Pooling layer after the convolutional part invariant to the input size.
By modifying the param file accordingly you will be able to manage smaller windows.