DecisionTreeRegressor and DecisionTreeClassifier do not allow `n_subfeatures=-1`.

[john-waczak]

The docs for DecisionTreeClassifier hyperparameter :n_subfeatures state:

n_subfeatures=0: number of features to select at random (0 for all, -1 for square root of number of features)

However the value -1 leads to the error "number of features -1 must be >= zero ". I'm not sure if the issue should be handled at this interface or in DecisionTree.jl itself as the function build_forest() has

if n_subfeatures == -1
    n_features = size(features, 2)
    n_subfeatures = round(Int, sqrt(n_features))
 end

while the function build_tree() does not. See this link for the classifier and this other link for the regressor.

[ablaom]

I believe the square-root option is probably only useful in the case of forests. Generally I haven't found feature subsampling at nodes useful for single trees. In any case, as you observed, it is not provided as an option in DecisionTree.jl trees. Unless you have a use-case for this kind of regularisation in single trees, I suggest we regard this as an error in the docs.

[john-waczak]

That sounds good to me. I'll update the docs.