Population Stability Index

Calculate the PSI between the features in two dataframes, X1 and X2.

X1 and X2 should be bucketed (outputs of fitted bucketers).

\[ PSI = \sum((\%{ Good } - \%{ Bad }) imes \ln rac{\%{ Good }}{\%{ Bad }}) \]

Parameters:

Name	Type	Description	Default
X1	DataFrame	bucketed features, expected	required
X2	DataFrame	bucketed features, actual data	required
epsilon	float	Amount to be added to relative counts in order to avoid division by zero in the WOE calculation.	0.0001
digits		(int): number of significant decimal digits in the IV calculation	None

Examples:

from skorecard import datasets
from sklearn.model_selection import train_test_split
from skorecard.bucketers import DecisionTreeBucketer
from skorecard.reporting import psi

X, y = datasets.load_uci_credit_card(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(
    X,y,
    test_size=0.25,
    random_state=42
)

dbt = DecisionTreeBucketer()
X_train_bins = dbt.fit_transform(X_train,y_train)
X_test_bins = dbt.transform(X_test)

psi_dict = psi(X_train_bins, X_test_bins)

Source code in skorecard/reporting/report.py

def psi(X1: pd.DataFrame, X2: pd.DataFrame, epsilon=0.0001, digits=None) -> Dict:
    """
    Calculate the PSI between the features in two dataframes, `X1` and `X2`.

    `X1` and `X2` should be bucketed (outputs of fitted bucketers).

    $$
    PSI = \\sum((\\%{ Good } - \\%{ Bad }) \times \\ln \frac{\\%{ Good }}{\\%{ Bad }})
    $$

    Args:
        X1 (pd.DataFrame): bucketed features, expected
        X2 (pd.DataFrame): bucketed features, actual data
        epsilon (float): Amount to be added to relative counts in order to avoid division by zero in the WOE
            calculation.
        digits: (int): number of significant decimal digits in the IV calculation

    Returns: dictionary of psi values. keys are feature names, values are the psi values

    Examples:

    ```python
    from skorecard import datasets
    from sklearn.model_selection import train_test_split
    from skorecard.bucketers import DecisionTreeBucketer
    from skorecard.reporting import psi

    X, y = datasets.load_uci_credit_card(return_X_y=True)
    X_train, X_test, y_train, y_test = train_test_split(
        X,y,
        test_size=0.25,
        random_state=42
    )

    dbt = DecisionTreeBucketer()
    X_train_bins = dbt.fit_transform(X_train,y_train)
    X_test_bins = dbt.transform(X_test)

    psi_dict = psi(X_train_bins, X_test_bins)
    ```
    """  # noqa
    assert (X1.columns == X2.columns).all(), "X1 and X2 must have same columns"

    y1 = pd.Series(0, index=X1.index)
    y2 = pd.Series(1, index=X2.index)

    X = pd.concat([X1, X2], axis=0)
    y = pd.concat([y1, y2], axis=0).reset_index(drop=True)

    psis = {col: _IV_score(y, X[col], epsilon=epsilon, digits=digits) for col in X1.columns}

    return psis