Adaboost: Remove deprecated argument 'algorithm'

[janezd]

Issue

In Sklearn 1.6, the only acceptable value for AdaBoost's argument algorithmisSAMME`. In 1.8, the argument will be removed. (scikit-learn/scikit-learn#29997)

Description of changes

• I removed the argument. This doesn't break any tests, so it apparently didn't have any effect (or we didn't test it properly). Any potential changes are however unavoidable because the current algorithm is removed from SkLearn (for the reasons, see RFC deprecate the SAMME.R algorithm in AdaBoostClassifier scikit-learn/scikit-learn#26784).
• I removed the setting from the widget. I haven't provided any migration, not even showing a warning because the user would have no way to disable the warning. Besides, it probably doesn't matter.
• Some tests depend on random seed. I change one of them to 4 so that results in scikit-learn=1.6.dev are the same as in earlier version(s).

Includes

• Code changes
• Tests
• Documentation

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 88.41%. Comparing base (9497b39) to head (27a6c61).
Report is 4 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #6910      +/-   ##
==========================================
+ Coverage   88.40%   88.41%   +0.01%     
==========================================
  Files         326      326              
  Lines       71971    71974       +3     
==========================================
+ Hits        63624    63636      +12     
+ Misses       8347     8338       -9     

[janezd]

If random seed is 0, the result today is not the same as it was the last week. I discovered the 4 seems OK.

Scikit and pandas are moving targets. :(

[VesnaT]

With the a new version of scikit-learn a default value of unsettable (within Orange) parameter will change (algorithm).

Should we set the algorithm parameter to SAMME to assure the same behaviour as when the scikit 1.6 is released?

[VesnaT]

box variable is unused.

[janezd]

@VesnaT I put the algorithm argument back, with the default value of SAMME. As I understand, our wrapper for scikit learners will skip it when scikit-learn will no longer accept it (in version 1.8).

<rant>In the meantime, we're back in red because scikit developers decided it would be a terribly good idea to change the behaviour of AUC (scikit-learn/scikit-learn#27412). But this is something for the next PR...

Why do we depend on unstable packages like pandas, sklearn and scipy? No better alternatives? Shame, really.

Sorry, I'm pissed. I thought I'd be doing something useful today, not hunting another change in sklearn.</rant>