#label-shift

Question 1: Single choice

A pneumonia classifier was trained on chest X-rays from a hospital where 10% of patients have pneumonia. It is deployed at a clinic during flu season, where 40% do. How pneumonia looks on an X-ray is the same in both places. Which kind of distribution shift is this, and what is the matching correction?

Label shift; reweight by $p(y)/q(y)$ , estimating the new $p(y)$ with a confusion matrix1
Concept shift; relabel the clinic's data, since the definition of pneumonia has changed2
Covariate shift; reweight by $p(\mathbf{x})/q(\mathbf{x})$ estimated with a domain classifier3
No shift, because each class looks the same in both places, so no correction is needed4

Timed test