type

assignment.SoftAssignment

class SoftAssignment(Assignment):

Differentiable linear-assignment solver using Sinkhorn iterations.

Computes an entropy-regularized optimal-transport plan for the cost matrix (via log-domain Sinkhorn iterations) and extracts a discrete assignment by taking mutual-argmax pairs over the plan. The plan itself is fully differentiable; use :func:`.sinkhorn_log_plan` directly when training with a soft-assignment loss.

Members

method

  • __init__Initialize the soft-assignment module.
  • solve_with_planSolve and return matches together with the Sinkhorn log-plan.

Source: unitrack/assignment/_soft.py:29