Implement proper feature scaling

[mef51]

For each feature x (e.g. data point in the ACF) we can rescale so x = (x - np.mean(corr) / np.std(corr)) to improve convergence speed, likely dramatically.