The hardware and bandwidth for this mirror is donated by METANET, the Webhosting and Full Service-Cloud Provider.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]metanet.ch.

Coreset 1.0.0 (2026-09-09)

Coreset selects a representative subset of a fixed candidate set under an arbitrary distance, solving four discrete location objectives on a distance matrix, a coordinate matrix, or an on-demand distance-column oracle.

Max-Min diversity (MMDP / discrete p-dispersion)

Maximises the minimum pairwise distance within a subset of size k.

• FarFirst(): greedy farthest-first selection (Gonzalez 1985), with a choice of peripheral seeding strategies, distinct-seed random restarts (nSeeds), and a robust ensemble default.
• DropAdd(): DropAdd tabu search (Porumbel et al. 2011), which can compute distances between pairs on the fly rather than needing a complete matrix a priori.
• Grasp(): GRASP with path relinking (Resende et al. 2010), attaining the highest T_k of the package’s heuristics on small to medium instances.
• ExactMaxMin(): exact node-packing optimum (Sayyady & Fathi 2016), decided by clique search.

Max-Mean dispersion

• MaxMean(): reinforcement-learning-guided tabu search (Nijimbere et al. 2020), selecting a subset of unrestricted size that maximises the mean pairwise distance.

Discrete k-centre

Minimises the largest distance from any element to its nearest selected centre.

• KCentre(): the CDSh covering heuristic (Garcia-Diaz et al. 2017, 2019).
• ExactKCentre(): exact minimum-cover optimum.

Max-Sum diversity and maximum entropy

• ExactMaxSum(): exact solver for the Max-Sum Diversity Problem (requires ‘highs’).
• MaxEntropy(): maximum-entropy (maxdet) selection — the mode of a determinantal point process — by greedy pivoted-Cholesky selection, and by exact enumeration for small instances.

Scoring and utilities

• MinDist(), MeanDist() and KCentreRadius() score an arbitrary selection under the max-min, max-mean and k-centre objectives respectively.
• PickPoint() exposes the peripheral seed indices directly.
• DropAdd() and Grasp() accept a maxCandidates composable-coreset cap, thinning the candidate set with FarFirst() before the expensive search and mapping the chosen indices back to the original numbering.
• Each solver returns a classed object with print(), format() and (where informative) summary() methods giving a terse or detailed report of the selection, the achieved objective, and the search effort.
• Solver behaviour is tunable through options(Coreset.symmetryTolerance = ), which sets how large a rounding discrepancy between d[i, j] and d[j, i] is repaired rather than refused, and options(Coreset.progress = ).

These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.