ce-calibrated-predict

Calibrated predictions with uncertainty bounds are generated by passing your data through a fully fitted WrapCalibratedExplainer, which validates inputs and returns point predictions or probabilities with optional two-sided intervals.

The predict method provides calibrated point predictions for regression or classification, while predict_proba returns calibrated probabilities for classification tasks, both supporting optional uncertainty intervals to reflect model confidence.

Threshold-based probabilities and Mondrian-group conditioning are supported for calibrated predictions, allowing flexible usage and integration when generating calibrated probabilities for classification and regression tasks.

A fully fitted and calibrated WrapCalibratedExplainer is required before generating calibrated probabilities, as the Skill validates inputs against this explainer to return reliable predict and predict_proba outputs with uncertainty intervals.

Calibrated probabilities are necessary when model confidence is miscalibrated because they provide reliable outputs for decision-making, ensuring the predicted probabilities accurately reflect true likelihoods with quantifiable uncertainty bounds.

Calibrated predictions work for both classification and regression tasks, returning calibrated point predictions via predict or calibrated probabilities via predict_proba, with optional uncertainty bounds to reflect model confidence.