Embeddings searchresult

Development build. This page describes main, not a released package. The latest published Lodestar.Embeddings is 0.4.0 — read its documentation.

SearchResult

One search hit: where the vector sits in the index, and how well it scored.

public readonly record struct SearchResult(int Index, float Score)

Example — reading a hit apart.

using Lodestar.Embeddings.Search;

var index = new EmbeddingIndex(dimension: 2);
index.Add(new float[] { 1f, 0f }, "first");
index.Add(new float[] { 0f, 1f }, "second");

SearchResult hit = index.Search(new float[] { 1f, 0f }, k: 1)[0];
int where = hit.Index;  // => 0
string id = index.GetId(hit.Index)!;  // => first

Remarks — two fields and nothing else.

Index is a position in [0, Count), in insertion order — the n-th vector added is index n − 1. It is what EmbeddingIndex.GetId takes, and what you carry back to whatever holds your documents.

Score is the similarity, which for a normalizing index is a cosine in [-1, 1]: 1 is the same direction, 0 perpendicular. For an index built with normalize: false it is a raw dot product and has no bounded range.

The document is deliberately not in here. This struct is 8 bytes with no references, so Search can score into a SearchResult[Count] and sort it without the garbage collector ever scanning that array or the sort moving references through it. Putting a string id in it would cost that on every query, to save one lookup on the handful of results actually returned.

Being a record struct, it compares by value and deconstructs — var (position, score) = hit; works.

Applies to — net10.0, netstandard2.0.

See also — EmbeddingIndex.Search, EmbeddingIndex.GetId, the search index.