Skip to content

Missing Categorical feature supports for HistgradientBoost #74

@yl4070

Description

@yl4070

HistgradientBoost methods from Scikit-learn support native categorical features, meaning no preprocessing is needed, as shown here:
https://scikit-learn.org/stable/modules/ensemble.html#categorical-support-gbdt

However, the MLJ interface seems to enforce the input to be continuous tables, as seen in the source code:

meta(HistGradientBoostingClassifier,
    input   = Table(Continuous),
    target  = AbstractVector{<:Finite},
    weights = false
    )

Since MLJ enforces the scitype schema, the categorical feature columns should be auto-inferred. I hope this can be addressed, thanks.

On second thoughts, since scikit-learn will auto-infer categorical features based on dtype, maybe relaxing the Table type to some union type would suffice.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions