torchcell.losses.multi_dim_nan_tolerant.isotonic_l2_pav

isotonic_l2_pav(y: Tensor, weights: Tensor | None = None) → Tensor

Solve isotonic regression using the PAV (Pool Adjacent Violators) algorithm.

Finds argmin_{v_1 >= v_2 >= … >= v_n} 0.5 * ||v - y||^2_2

Parameters:

• y – Input values (1D tensor)

• weights – Optional weights (1D tensor)

Returns:

Solution to isotonic regression