Random Forest classifier does not work correctly with scikit >= 1.4.1

[vkozmik]

Connected with the change in scikit, I think that the values of "votes" should be divided by "node_count"
API Change The tree_.value attribute in tree.DecisionTreeClassifier, tree.DecisionTreeRegressor, tree.ExtraTreeClassifier and tree.ExtraTreeRegressor changed from an weighted absolute count of number of samples to a weighted fraction of the total number of samples. #27639 by Samuel Ronsin.

Suggest to change line 276 of tree.py:

Originally:
votes = [[[float(votes_for(c).get('recordCount')) if votes_for(c) is not None else 0.0 for c in classes]]]

Suggested:
votes = [[[float(votes_for(c).get('recordCount')) / node_count if votes_for(c) is not None else 0.0 for c in classes]]]

[iamDecode]

Great, thanks for figuring this out! I would have to check if changing it to a weighted fraction is still compatible with older versions of scikit-learn.

[vkozmik]

You're welcome. Alternatively, we can wrap the new code in some condition for scikit version

major, minor, patch = map(int, sklearn.version.split("."))
if major <= 1 and (minor <= 3 or (major == 4 and patch ==0)):
old code
else
new code

not nice, but probably solves the issue