What is Max Absolute Scaler? Compare it with MinMax Normalization? Why scaling to [-1, 1] might be better than [0, 1] scaling?