fieldopt-core 0.10.0

Field-survey design under real field cost: travel matrices, GRASP routing of field work (TSP/VRP), spatially balanced sampling (local pivotal method), Horvitz-Thompson estimation with local-mean variance, and cost-aware allocation. Pure Rust, no dependencies.
Documentation

fieldopt-core

Field-survey design under real field cost, in pure Rust with no dependencies:

  • Travel matrices from coordinates (great-circle or Euclidean), or supplied by the user from a road network (for example OSRM).
  • Routing of field work: instances of up to 13 units are solved exactly (dynamic programming over subsets); larger ones by a hybrid genetic search (Vidal 2022) with feasible and infeasible subpopulations under adaptive penalties, the optimal split of Prins (2004), and a granular local search (2-opt, Or-opt, relocate, swap, SWAP*, 2-opt*) priced from route totals in constant time, under limits on route duration (travel plus service time), stops and capacity, with one or several depots (each with its own number of routes available) and a fixed cost per route, on symmetric or asymmetric matrices. Optimal on the seven classical TSPLIB instances up to 101 nodes and on 24 of the 27 Augerat set A CVRP instances (mean gap 0.01%, largest 0.15%) in about two seconds each; on Cordeau's 23 multi-depot instances (50 to 360 customers, with route duration limits) at or below the best known solution on 13, mean gap 0.20%, largest 1.6%, in 1 to 100 seconds; results vary by less than one percent across seeds. A branch-and-bound on Held-Karp 1-trees proves single tours optimal (six of the seven TSPLIB instances within ten seconds); stops on offspring or on time.
  • Spatially balanced sampling (a grid index makes the local pivotal method and the spatial balance measure near linear: a million units in under a second) with the local pivotal method (Grafström, Lundström and Schelin 2012) and size-proportional inclusion probabilities.
  • Design-based estimation: Horvitz-Thompson total with the local-mean variance estimator (Grafström and Schelin 2014) over each unit and its k nearest sampled units (identical to BalancedSampling::vsb()), and the simple-random-sampling variance as a reference.
  • Cost-aware allocation across strata or frames: minimum cost for a target variance, or minimum variance for a budget (Cochran 1977).

Every randomised step is reproducible from a seed. The crate is the engine of the R package fieldopt (CAST Lab, Universidade Federal de Pernambuco) and can be used on its own.

[dependencies]
fieldopt-core = "0.1"

See the crate documentation for an end-to-end example (select segments, route the field work, estimate).

Version 0.2 adds the pieces an area-frame or dual-frame survey needs:

  • spatial: order of units along a Hilbert curve and systematic PPS sampling along it, as independent interpenetrating replicates for design-based variance estimation.
  • dualframe: Hartley's dual-frame allocation (area + list with overlap): cost-optimal sample sizes in each frame and the optimal mixing weight theta, or the screening design (theta = 0), for a target variance or a budget, with design effects.
  • twostage: Cochran's two-stage allocation with the cost function c1 n + c2 n m: optimal secondary units per primary unit and the number of primary units for a target variance or a budget.

License

GPL-3.0-or-later. © CAST Lab, Universidade Federal de Pernambuco.