fatf.transparency.models.feature_influence.partial_dependence

fatf.transparency.models.feature_influence.partial_dependence(dataset: numpy.ndarray, model: object, feature_index: Union[int, str], treat_as_categorical: Optional[bool] = None, steps_number: Optional[int] = None, include_rows: Union[int, List[int], None] = None, exclude_rows: Union[int, List[int], None] = None) → Tuple[numpy.ndarray, numpy.ndarray]

Calculates Partial Dependence for a selected feature.

Partial Dependence [FRIEDMAN2001GREEDY] is computed as a mean value of Individual Conditional Expectations (c.f. fatf.transparency.models.feature_influence.individual_conditional_expectation) over all the selected rows in the input dataset.

The input parameters, exceptions and warnings match those used in fatf.transparency.models.feature_influence.individual_conditional_expectation function.

Note

If you wish to have access to both ICE and PDP results consider using fatf.transparency.models.feature_influence.individual_conditional_expectation and fatf.transparency.models.feature_influence.partial_dependence_ice functions to minimise the computational cost.

FRIEDMAN2001GREEDY

J. H. Friedman. Greedy function approximation: A gradient boosting machine. The Annals of Statistics, 29:1189–1232, 2001. URL https://projecteuclid.org/euclid.aos/1013203451. [p421, 428]

Returns

partial_dependence_arraynumpy.ndarray

A 2-dimensional array of (steps_number, n_classes) shape representing Partial Dependence for all of the classes for selected rows (data points).

feature_linespacenumpy.ndarray

A one-dimensional array – (steps_number, ) – with the values for which the selected feature was substituted when the dataset was evaluated with the specified model.

Examples using fatf.transparency.models.feature_influence.partial_dependence

Using Partial Dependence Explainer