Apply hierarchy constraints via max-pooling over descendants (MCM)

Given neural network outputs x and ancestor matrix R, enforces that if a class is predicted positive, all its ancestors must also be positive. Implements: final_out[i] = max{x[j] : R[i,j] = 1}

Usage

get_constr_output(x, R)

Arguments

x

A torch_tensor of shape (batch_size, n_classes).

R

A torch_tensor of shape (1, n_classes, n_classes) where R[1, i, j] = 1 iff class i is a descendant of class j.

Value

A torch_tensor of shape (batch_size, n_classes) with constrained outputs.