Machine Learning Idea – The Model Predictions Must Be Exact For A True Low Poly Version Of The Signal. Found Out That You ?Must Invert One Of Two -F.relu()

A strategy for getting the model just right could be to sample (not resample) the signal to a low number of true samples of the original signal.

This way when you increase the samples you immediately see where the problem arises in the prediction.

From this I figured out that you can try with great success to invert one F.relu(). That is.

// Per Lindholm