Combat wrapper

fyt.wrappers.combat_wrapper

CombatWrapped

Bases: BaseEstimator, TransformerMixin

A wrapper around pycombat.Combat to handle polars/pandas DataFrame input and output.

This wrapper extends pycombat.Combat to be compatible with scikit-learn pipelines. Accepts polars DataFrames as input, converts internally to pandas for pycombat, and returns polars DataFrames as output.

The wrapper extracts the batch column from the DataFrame during fit and transform, and applies ComBat batch correction to the remaining numerical features. The underlying Combat model always runs in parametric mode (mode="p").

Parameters:

Name	Type	Description	Default
batch_column	str	Name of the column containing batch identifiers.	required
feature_columns	list[str] | None	Optional list of feature column names to apply Combat to. If None, all numerical columns except batch_column are used.	None
drop_batch_column	bool	If True, the batch column is dropped from the output DataFrame. If False (default), the batch column is kept.	False
conv	float	Convergence criterion for EB optimization (default: 0.0001).	0.0001

__getattr__(name)

Delegate attribute access to the underlying Combat instance.

Parameters:

Name	Type	Description	Default
name	str	Attribute name.	required

Returns:

Type	Description
	Attribute value from the underlying Combat instance.

Raises:

Type	Description
AttributeError	If the attribute is not found on the wrapper or the underlying Combat instance.

fit(X, y=None)

Fit the ComBat model.

Parameters:

Name	Type	Description	Default
X	DataFrame | DataFrame | ndarray	polars or pandas DataFrame with features and batch column.	required
y	object	Ignored, present for sklearn compatibility.	None

Returns:

Name	Type	Description
self	CombatWrapped	Fitted transformer.

Raises:

Type	Description
ValueError	If the batch column is not found in the input DataFrame.

get_feature_names_out(input_features=None)

Return output feature names, honoring the dropped batch column.

transform returns the input columns in their original order with the feature columns corrected, dropping the batch column when drop_batch_column is set. The output feature names mirror that.

Parameters:

Name	Type	Description	Default
input_features	Sequence[str] | ndarray | None	Names of the input columns. If None, the columns seen during fit are used.	None

Returns:

Type	Description
ndarray	numpy.ndarray of output feature names.

Raises:

Type	Description
RuntimeError	If the transformer is not fitted and no input_features were provided.

set_output(*, transform=None)

Set output configuration for compatibility with sklearn pipelines.

Parameters:

Name	Type	Description	Default
transform	str | None	Output format for transform method. Not used in this wrapper.	None

Returns:

Name	Type	Description
CombatWrapped	CombatWrapped	The transformer itself.

transform(X)

Transform the data using fitted ComBat parameters.

Parameters:

Name	Type	Description	Default
X	DataFrame | DataFrame	polars or pandas DataFrame with features and batch column.	required

Returns:

Name	Type	Description
X_out		polars DataFrame with ComBat-corrected features.

Raises:

Type	Description
RuntimeError	If the transformer has not been fitted yet.
ValueError	If the batch column or a fitted feature column is not found in the input DataFrame.