Convert true binary labels into PU labels

Turn any labeled binary dataset into a positive-unlabeled one under the SCAR assumption: each positive stays labeled with constant probability label_propensity and the rest become unlabeled. This is how the PU literature builds benchmarks from open binary datasets — positives of the known class remain 1, hidden positives and negatives pool as 0.

Usage

pu_mask(z, label_propensity = 0.5, seed = NULL)

Arguments

z

Binary vector of true labels (1 positive, 0 negative).

label_propensity

Constant probability that a true positive is observed as a labeled positive, in (0, 1].

seed

Optional nonnegative integer seed. Use a different seed than the one that generated z — reusing the same stream can correlate the mask with the simulated labels.

Value

An integer vector of PU labels (1 labeled positive, 0 unlabeled) the same length as z.

See also

simulate_pu_data(), xplus()

Examples

z <- c(rep(1, 80), rep(0, 120))
y <- pu_mask(z, label_propensity = 0.5, seed = 1)
fit <- xplus(matrix(rnorm(200 * 4), ncol = 4), y, max_iter = 3)