Linear least squares fitting - choosing the right algorithm

[thomasaarholt]

I'm struggling trying to choose the right algorithm to incorporate into Hyperspy to do MLLS fitting. Currently I'm using LinearRegression from scikit-learn. Unfortunately, the residual returned by its methods is being deprecated. I've raised an issue about this on their github.

Alternatives include numpy's lstsq and scipy's lstsq.

I'm basically trying to find the easiest way to do this.

[francisco-dlp]

Personally I would simply pick scipy's lstsq because sklearn's regression is just a wrapper around it. For something so simple, I think that we'll be better off getting it from the source.

We could also take it from numpy, but usually functions that are available from both scipy and numpy tend to be faster in scipy.

[tjof2]

A relevant thread: http://stackoverflow.com/questions/29372559/what-is-the-difference-between-numpy-linalg-lstsq-and-scipy-linalg-lstsq

But it looks like scipy now uses xGELSD by default anyway - plus offers a method option to change the LAPACK routine anyway,

[tjof2]

See #1321 #1462 #2422

[tjof2]

I think you can close this @thomasaarholt given #2422 ?

[thomasaarholt]

4 years later :p