Closed Flaminietta closed 1 year ago
From the first to the second figure, the only thing changing is the rescaling before estimating the mean and std. Therefore, data density should be the same.
Fine to me. About the question, I'm not sure why this is happening. Let me think a bit more about it
I don't understand the behavior and think we should keep an eye on it (maybe also after merging). Actually, I'd expect a contrary behavior: if - by rescaling between 0 and 1 - we are reducing the range where the data is spread, I'd then expect a smaller standard deviation.
This PR arises from my wish to be able to plot not only the rough scores but also the normalized ones also in the case some randomness is added to the weights. If randomness is added to the weights, this means that MCD has N-input matrices and produces N-outputs.
The logic I have implemented is:
This approach should retain the statistical properties of the rescaled values.