PercentileModelCalibration

Creates percentile mapping from model scores using ROC curve analysis for consistent risk interpretation

SageMaker step type	Processing
Node type	internal (consumes upstream, produces downstream)
Container entry point	percentile_model_calibration.py
Interface file	steps/interfaces/percentile_model_calibration.step.yaml

Compute

Compute kind	framework
SDK class	SKLearn (SageMaker DLC via image_uris.retrieve)

Functionality

Percentile model calibration that converts raw model scores to calibrated percentile values using ROC curve analysis. Supports single-task and multi-task calibration with configurable calibration dictionary.

Inputs (dependencies)

Input	Type	Required	Compatible producers
evaluation_data	processing_output	yes	XGBoostTraining, XGBoostModelEval, XGBoostModelInference, LightGBMTraining, LightGBMModelEval, LightGBMModelInference, LightGBMMTTraining, LightGBMMTModelEval, PyTorchTraining, PyTorchModelEval, PyTorchModelInference, ModelEvaluation, TrainingEvaluation, CrossValidation, ModelCalibration
calibration_config	processing_output	no	ConfigurationStep, DataPreprocessing, FeatureEngineering, ModelConfiguration

Outputs

Output	Type
calibration_output	processing_output
metrics_output	processing_output
calibrated_data	processing_output

Consumers (downstream steps)

Steps that declare this step as a compatible input source:

• Package

Framework requirements

Package	Version
scikit-learn	>=0.23.2,<1.0.0
pandas	>=1.2.0,<2.0.0
numpy	>=1.20.0

---

← Back to the Step Catalog