Features¶

Beyond the basic min-cost assignment, fastlap ships a set of features aimed squarely at real production workloads — tracking pipelines, scheduling systems, and large batch jobs.

Cost Limit (Gating)

Reject assignments that exceed (or fall below) a threshold cost — the gating step every multi-object tracker needs.
Batch Solving

Solve hundreds of independent matrices in parallel across all CPU cores via Rayon.
Weighted Costs

Multiply each cost entry by a per-element weight before solving, while total cost is still reported unweighted.
Optimal Duals

Solve and return the row/column dual potentials (u, v) — feasible, tight on the matching, with sum(u) + sum(v) equal to the optimum.
K-Best (Murty)

Rank the top-K alternative assignments in increasing cost order — for multi-hypothesis tracking.
Bottleneck (LBAP)

Minimise the maximum edge cost in the assignment, not the sum — the bottleneck assignment problem.
Sparse Matrices

Feed a scipy.sparse.csr_matrix straight into LAPMOD or LAPJVsp without ever densifying it.
Compatibility Layers

Drop-in replacements for scipy.optimize.linear_sum_assignment and lap.lapjv / lapx.lapjv — plus lapx-style lapjvx and assignment_pairs helpers.
Visualisation & Demos

Terminal heatmaps, an algorithm head-to-head, a bipartite-graph render, and a matplotlib overlay — runnable examples under examples/.