SmallWorldParameters Class

Definition

public class SmallWorldParameters

The single configuration object passed to a SmallWorld<TItem, TDistance>. Most defaults are sensible; the properties below control graph degree, build quality, query-time recall, and optional ACORN-γ filtering.

Remarks

Pick M first, build once with the defaults, then tune EfSearch at query time — it is the cheapest dial and gives a direct recall/latency curve without rebuilding. Construction knobs (M, LevelLambda, ConstructionPruning, the heuristic and ACORN settings) take effect at build time and require a rebuild to change. See Parameters.

Properties

M

public int M { get; set; }

Graph degree. Default 10. Higher values mean better recall and faster search, at the cost of memory and build time. Typical range 8–64; 16 is a reasonable default for float[] embeddings of dimension 100–1000.

LevelLambda

public double LevelLambda { get; set; }

Controls the level distribution. Default 1 / Math.Log(M). Keep in sync with M — set both together.

ConstructionPruning

public int ConstructionPruning { get; set; }

Number of candidates considered when inserting an item (efConstruction). Default 200. Larger values produce a higher-quality graph at the cost of build time, and do not affect query time.

NeighbourHeuristic

public NeighbourSelectionHeuristic NeighbourHeuristic { get; set; }

Strategy for picking neighbours during insertion. Default SelectSimple. SelectHeuristic keeps more diverse neighbours — useful for clustered data.

ExpandBestSelection

public bool ExpandBestSelection { get; set; }

Default false. Only meaningful with SelectHeuristic. Expands the candidate pool before pruning.

KeepPrunedConnections

public bool KeepPrunedConnections { get; set; }

Default false. Only meaningful with SelectHeuristic. Keeps pruned candidates as second-best edges.

EnableDistanceCacheForConstruction

public bool EnableDistanceCacheForConstruction { get; set; }

Default true. Caches distance computations during build. Disable only if distances are nearly free (for example a precomputed table lookup).

InitialDistanceCacheSize

public int InitialDistanceCacheSize { get; set; }

Initial size of the distance cache. Default 1_048_576. Reset to 0 on DeserializeGraph.

InitialItemsSize

public int InitialItemsSize { get; set; }

Pre-allocates the items list. Default 1024. Set to your expected dataset size to avoid resizing. Affects allocation strategy only — it does not pre-reserve graph nodes.

EfSearch

public int EfSearch { get; set; }

Number of candidates kept during the query-time graph walk. Default 50. Higher values mean better recall at the cost of latency. This is the runtime dial — change it between queries without rebuilding the graph.

OptimizeForFiltering

public bool OptimizeForFiltering { get; set; }

Default false. Enables the ACORN-γ construction strategy for filtered nearest-neighbor search. Must be set before calling AddItems. See Filtering.

Gamma

public int Gamma { get; set; }

ACORN neighbour expansion factor (γ). Default 1. Higher values build a denser graph that survives more selective filters at the cost of memory and build time; 3–5 is a reasonable starting range.

Mb

public int Mb { get; set; }

ACORN compression parameter for layer 0. Default 10. Higher values make layer 0 less dense and recover memory at the cost of recall under filters.

Applies to