mapie/deprecated-in-mapie
| Rule | deprecated-in-mapie |
|---|---|
| Pack | MAPIE |
| Severity | warning |
| Fix | safe, with xpd check --fix |
| Library versions | MAPIE 1.3 and later |
MAPIE renamed or stopped reading these forms, and they warn. Use the replacement that the message names.
risk = BinaryClassificationRisk(risk_occurrence=missed, risk_condition=positive, higher_is_better=False)y_pred, y_pis = cross.predict_interval(X_test, aggregate_predictions="median")series.update(X_new, y_new, gamma=0.05)risk = BinaryRisk(risk_occurrence=missed, risk_condition=positive, higher_is_better=False)y_pred, y_pis = cross.predict_interval(X_test, aggregate_point_predictions="median")series.update(X_new, y_new)series.adapt_conformal_inference(X_new, y_new, gamma=0.05)What the rule reports
Section titled “What the rule reports”The rule reports forms that MAPIE renamed or stopped reading:
| Old form | Since | Write instead |
|---|---|---|
BinaryClassificationRisk |
1.4 | BinaryRisk |
aggregate_predictions= on CrossConformalRegressor.predict and predict_interval |
1.4.1 | aggregate_point_predictions= |
ensemble= on the same methods of JackknifeAfterBootstrapRegressor |
1.4.1 | aggregate_point_predictions= |
gamma=, confidence_level= and optimize_beta= on TimeSeriesRegressor.update, which reads none of them |
1.3 | update without them, and adapt_conformal_inference for gamma |
How to fix it
Section titled “How to fix it”xpd check --fix applies the fix of this rule.
The safe fix writes BinaryRisk for BinaryClassificationRisk, and aggregate_point_predictions= for aggregate_predictions= and for ensemble= on a JackknifeAfterBootstrapRegressor. update has no fix, because the call of adapt_conformal_inference is a new statement.
The fixes are safe. A controller built with BinaryRisk picked the same thresholds and gave the
same predictions as one built with BinaryClassificationRisk, and the intervals with the new
keywords were bit equal to the ones with the old keywords. A file that imports both risk classes
ends with BinaryRisk twice in its import.
ensemble= is reported only on an object that the file assigns from
JackknifeAfterBootstrapRegressor and from no other class, because TimeSeriesRegressor.predict
still takes it. With method="aci", the intervals after
update(X, y, confidence_level=0.9, gamma=0.5) were bit equal to the ones after
update(X, y, confidence_level=0.9) on MAPIE 1.3.0, 1.4.0 and 1.6.0. They differed on MAPIE
1.2.0, which needs confidence_level in update, so the update document carries since: "1.3".
Message
Section titled “Message”xpd prints this message with the issue. A name that starts with $ stands for code from your file.
BinaryClassificationRisk is deprecated since MAPIE 1.4 and raises FutureWarning. Use BinaryRisk, which takes the same arguments and gives the same result.Browse the other rules of the MAPIE pack.