scikit-learn Integration¶

SWIFTMonitor inherits from sklearn.base.BaseEstimator and sklearn.base.TransformerMixin, making it a first-class scikit-learn citizen.

Estimator API¶

Parameter Management¶

from swift import SWIFTMonitor

monitor = SWIFTMonitor(model=model, n_permutations=200, alpha=0.05)

# Inspect all parameters
monitor.get_params()
# {'model': ..., 'order': 1, 'n_permutations': 200, 'alpha': 0.05,
#  'correction': 'benjamini-hochberg', 'n_synthetic': 10,
#  'max_samples': None, 'random_state': 42}

# Modify parameters
monitor.set_params(alpha=0.01, correction="bonferroni")

Cloning¶

Cloning creates a new unfitted estimator with identical parameters:

from sklearn.base import clone

monitor2 = clone(monitor)
# monitor2 is unfitted, with same parameters as monitor

Transformer API¶

fit_transform¶

The standard fit_transform shorthand works as expected:

# These are equivalent:
X_transformed = monitor.fit_transform(X_ref)

# and:
monitor.fit(X_ref)
X_transformed = monitor.transform(X_ref)

transform() maps each observation to its bucket's mean SHAP value. The output has the same shape and column names as the input.

Sample vs. Sample Comparison¶

By default, score() and test() compare monitoring data against the fitted reference. Pass X_compare to compare two arbitrary samples instead — the SHAP transformation (buckets + mean SHAP) is always the one learned from the reference:

# Score: monitoring vs. drifted
scores = monitor.score(X_mon, X_compare=X_drifted)

# Full test: monitoring vs. drifted
result = monitor.test(X_mon, X_compare=X_drifted)
print(result.drifted_features)

This is useful when you want to compare two time windows of production data against each other, rather than against the original reference.