Add CaliForest model, data loader, task, and calibration metrics

[cgigliob]

CaliForest: Calibrated Random Forest for Health Data

Contributors:

• Tony L. (tonyln2@illinois.edu)
• Cesar G. (cesarjg2@illinois.edu)
• Arjun T. (avtange2@illinois.edu)

Contribution Type: Full Pipeline (Dataset + Task + Model)

Paper: Y. Park and J. C. Ho. "CaliForest: Calibrated Random Forest for Health Data." ACM CHIL, 2020. https://doi.org/10.1145/3368555.3384461

Description:
Reimplementation of CaliForest within PyHealth. CaliForest uses out-of-bag prediction variance with an Inverse-Gamma prior to learn a calibrated random forest without requiring a held-out calibration set. Supports isotonic, logistic, and beta calibration. Includes dataset loader for MIMIC-Extract, parameterized task for 4 binary prediction targets, and 6 calibration metrics.

Files to review:

New files:

• pyhealth/models/califorest.py — CaliForest model (BaseModel)
• pyhealth/datasets/califorest_mimic_extract.py — MIMIC-Extract loader (BaseDataset)
• pyhealth/datasets/configs/califorest_mimic_extract.yaml — Dataset config
• pyhealth/tasks/mimic_extract_califorest.py — Task (BaseTask)
• pyhealth/metrics/califorest_calibration.py — 6 calibration metrics
• tests/test_califorest.py — Model tests (7)
• tests/test_calibration_metrics.py — Metrics tests (11)
• tests/test_mimic_extract_califorest_task.py — Task tests (7)
• tests/test_califorest_mimic_extract_dataset.py — Dataset tests (6)
• examples/califorest_mimic_extract_mortality_califorest.py — Ablation script
• docs/api/models/pyhealth.models.CaliForest.rst
• docs/api/datasets/pyhealth.datasets.CaliForestMIMICExtractDataset.rst
• docs/api/tasks/pyhealth.tasks.MIMICExtractCaliForestTask.rst

Modified files:

• docs/api/models.rst — Added toctree entry
• docs/api/datasets.rst — Added toctree entry
• docs/api/tasks.rst — Added toctree entry

31 tests passing. All use synthetic data.