example

Documentation

Index

• func FormatGroundTruth(neighbors []int, distances []float64, k, maxResults int) string
• func FormatResults(results []core.Neighbor, maxResults int) string
• func LoadCSV(index core.Index, path string, skipHeader bool) error
• func LoadDataset(index core.Index, dir string) (testVectors [][]float32, trueNeighbors [][]int, trueDistances [][]float64, ...)
• func LoadTestDataset(dir string) ([][]float32, [][]int, [][]float64, error)
• func LoadTrainingVectors(dir string) (map[int][]float32, error)
• func RecallAtK(predicted []core.Neighbor, groundTruth []int, k int) float64
• func RunDataset(factory IndexFactory, dataset, root string, k, numQueries, maxResults int)
• type IndexFactory
• type QueryResult

Functions

func FormatGroundTruth

func FormatGroundTruth(neighbors []int, distances []float64, k, maxResults int) string

FormatGroundTruth returns a formatted string of ground-truth neighbor results. maxResults specifies how many items to include.

func FormatResults

func FormatResults(results []core.Neighbor, maxResults int) string

FormatResults returns a formatted string of neighbor results. maxResults specifies how many items to include.

func LoadCSV

func LoadCSV(index core.Index, path string, skipHeader bool) error

LoadCSV reads float32 vectors from a CSV file and adds them to the index.

func LoadDataset

func LoadDataset(index core.Index, dir string) (
	testVectors [][]float32,
	trueNeighbors [][]int,
	trueDistances [][]float64,
	err error,
)

LoadDataset loads a dataset from a directory into the given index. The directory must contain the following files:

• train.csv (vectors to add to the index)
• test.csv (query vectors, not added to the index)
• neighbors.csv (expected neighbor IDs per query)
• distances.csv (expected distances per query)

func LoadTestDataset

func LoadTestDataset(dir string) ([][]float32, [][]int, [][]float64, error)

LoadTestDataset loads the test vectors and ground-truth data from the specified directory. It returns the test vectors, true neighbor IDs, and true distances (ground-truth).

func LoadTrainingVectors

func LoadTrainingVectors(dir string) (map[int][]float32, error)

LoadTrainingVectors loads training vectors from "train.csv" in the specified directory. It returns a map from id (row number, 0-indexed) to the vector.

func RecallAtK

func RecallAtK(predicted []core.Neighbor, groundTruth []int, k int) float64

RecallAtK computes Recall@k as the fraction of all ground-truth items that appear in the top k predictions.

func RunDataset

func RunDataset(factory IndexFactory, dataset, root string, k, numQueries, maxResults int)

RunDataset loads the dataset, builds the index using the provided factory, and runs kNN queries on a subset of test queries. If numQueries is negative or exceeds the number of available test vectors, all test vectors are used. It prints predicted results, ground-truth (if not benchmarking), and computes Recall@k along with per-query response times, average response time, and overall runtime. When benchmarking, a progress bar is displayed. The number of worker threads is read from the HANN_BENCH_NTRD environment variable.

Types

type IndexFactory

type IndexFactory func() core.Index

IndexFactory is a function that creates a new index.

type QueryResult

type QueryResult struct {
	// contains filtered or unexported fields
}

QueryResult holds the results for a single query.