NN-Aging: fix normalization

[talumbau]

Force normalization to be only one training set uniformly across notebook.

[talumbau]

Hey @rickecon @jdebacker I think this is a good fix for the NN training. Here is what I get with the fix:

I changed the normalization function like this:

def normalize_output(data, data_max=None, data_min=None):
    """Normalizes the output data to a [0, 1] range."""
    if data_max is None:
        data_max = np.max(data)
    if data_min is None:
        data_min = np.min(data)
    return (data - data_min) / (data_max - data_min)

and then just force the max and min to be only the values from the training set partition. I think that was problem. Take a look and let me know what you think!

[talumbau]

Ha! I see this is already fixed in #2

[jdebacker]

@talumbau Yes, we finally caught this. But still having trouble with fit. Will send email update.