Estimates \(\theta, q_0, q_1\) by maximizing the joint likelihood built from
labeled calibration data and unlabeled surrogate outputs, then obtains
standard errors via the inverse Fisher information (observed and expected).
Usage
fit_misclass_mle(y_cal, yhat_cal, yhat_test, level = 0.9)
Arguments
- y_cal
Human labels on the calibration split (0/1).
- yhat_cal
Surrogate predictions on the calibration split (0/1).
- yhat_test
Surrogate predictions on the unlabeled/test split (0/1).
- level
Confidence level for the Wald intervals. Defaults to 0.90.
Value
A list containing the MLEs, variance-covariance matrices from the
observed and expected Fisher information, and Wald intervals for \theta.