Skip to content

Clarify that n_trees cannot use generated text features #608

Description

@ElouenGinat

Documentation issue

The n_trees parameter is currently documented as follows:

Maximum number of decision tree features to construct. The constructed trees combine other features, either native or constructed.

The phrase "either native or constructed" can reasonably be understood as including the text features generated by Khiops through n_text_features. However, tree construction cannot use these generated text features (this is about features extracted from Text/TextList variables, not the raw text variables themselves).

The Khiops GUI documentation states this limitation explicitly:

Constructed variables and text features are generated independently. Trees and pairs can exploit constructed variables, but not text features.

Reference: https://khiops.org/ui-docs/khiops/#feature-engineering

The Python API documentation does not currently expose this distinction clearly: https://khiops.org/api-docs/python-api/

Suggested change

Please clarify the n_trees parameter documentation for KhiopsClassifier, KhiopsRegressor, and KhiopsEncoder. For example:

Maximum number of decision tree features to construct. The constructed trees can combine native variables and automatically constructed variables, but they cannot use text features generated from Text or TextList variables.

This would prevent users from expecting n_trees to model interactions between Khiops-generated textual features.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions