Missing values processing

The missing values processing mode depends on the feature type.

Numerical features

t-boost interprets the value of a numerical feature as a missing value if it is equal to one of the following values:

• null in a polars column, or None
• Floating point NaN value

Positive and negative infinity are not missing values: they fall into the highest and the lowest bin of the feature.

Missing values are put into a reserved bin of their own. When a split on the feature is selected, the missing values are tried on both sides of the split, and the side with the better score is kept. In the rating tables, every numeric axis has a cell for missing values (cell 0), so missing values get a relativity of their own.

Categorical features

Missing values (null, None or NaN) of a categorical feature are collected into one reserved level, which is encoded like any other level. A missing value is never treated as an unseen value. The categories_ attribute lists the missing level as None when the fit saw missing values.