| """Local scaling from training residuals (expected to fail).""" |
| import numpy as np |
| from .base import ConformalResult |
| from ._split_quantile import split_conformal_quantile |
| from ._knn_sigma import knn_sigma_hat |
|
|
|
|
| def trainres_conformal( |
| R_cal: np.ndarray, |
| R_test: np.ndarray, |
| alpha: float, |
| U_cal: np.ndarray, |
| U_test: np.ndarray, |
| R_train: np.ndarray, |
| U_train: np.ndarray, |
| k: int = 20, |
| ) -> ConformalResult: |
| """Locally-normalized conformal using training residuals for scale. |
| |
| Args: |
| R_cal: calibration residuals (n_cal,) |
| R_test: test residuals (n_test,) |
| alpha: miscoverage level |
| U_cal: calibration predictions (n_cal, K) |
| U_test: test predictions (n_test, K) |
| R_train: training residuals (n_train,) |
| U_train: training predictions (n_train, K) |
| k: kNN neighbors for scale estimation |
| |
| Returns: |
| ConformalResult (expected to be miscalibrated). |
| """ |
| sigma_hat_cal = knn_sigma_hat(U_train, R_train, U_cal, k=k) |
| S_cal = R_cal / sigma_hat_cal |
| q = split_conformal_quantile(S_cal, alpha) |
|
|
| sigma_hat_test = knn_sigma_hat(U_train, R_train, U_test, k=k) |
| radius = sigma_hat_test * q |
| covered = R_test <= radius |
|
|
| return ConformalResult(covered=covered, radius=radius, threshold=q) |
|
|
