dataeval.core.balance_classwise¶

dataeval.core.balance_classwise(class_labels, factor_data, discrete_features=None, num_neighbors=5)¶

Mutual information (MI) between factors (class label, metadata, label/image properties).

Parameters:¶

class_labels : NDArray[np.intp]¶: Target class labels as integer indices.
factor_data : NDArray[np.intp]¶: Factor values after binning or digitization.
discrete_features : Iterable[bool] | None = None¶: Boolean array or iterable defining whether or not the feature set is discretized.
num_neighbors : int = 5¶: Number of points to consider as neighbors.

Returns:¶

(num_factors+1) x (num_factors+1) estimate of mutual information between num_factors metadata factors and class label. Symmetry is enforced.

Return type:¶

NDArray[np.float64]

Notes

We use mutual_info_classif from sklearn since class label is categorical. mutual_info_classif outputs are consistent up to O(1e-4) and depend on a random seed. MI is computed differently for categorical and continuous variables.

Example

Return balance (mutual information) of factors with class_labels

>>> metadata = generate_random_metadata(
...     labels=["doctor", "artist", "teacher"],
...     factors={
...         "age": [25, 30, 35, 45],
...         "income": [50000, 65000, 80000],
...         "gender": ["M", "F"]},
...     length=100,
...     random_seed=175)

Return classwise balance (mutual information) of factors with individual class_labels

>>> bal_cw = balance_classwise(
...     class_labels=metadata.class_labels,
...     factor_data=metadata.binned_data,
...     discrete_features=[True, True, True])
>>> bal_cw
array([[7.818e-01, 1.388e-02, 1.803e-03, 7.282e-04],
       [7.084e-01, 2.934e-02, 1.744e-02, 3.996e-03],
       [7.295e-01, 1.157e-02, 2.799e-02, 9.451e-04]])