Skip to content

Latest commit

 

History

History
107 lines (83 loc) · 4.15 KB

Cpp_API.rst

File metadata and controls

107 lines (83 loc) · 4.15 KB
Raw

C++ code example

#include "hnsw.h"
#include <vector>
#include <map>

int main() {
    n2::Hnsw index(3, "angular");
    index.AddData(std::vector<float>{0, 0, 1});
    index.AddData(std::vector<float>{0, 1, 0});
    index.AddData(std::vector<float>{0, 0, 1});
    std::vector<std::pair<std::string, std::string>> configs;
    int n_threads = 10;
    int M = 5;
    int MaxM0 = 10;
    index.Build(M, MaxM0, -1, n_threads);
    std::vector<std::pair<int, float> > result;
    int ef_search = 3*10;
    index.SearchByVector(std::vector<float>{3, 2, 1}, 3, ef_search, result);
    return 0;
}

C++ API

Note that if a user passes a negative value, it will be set to a default value when the metric has the default value.

  • n2::HnswIndex(int dim, std::string metric = "angular"): returns a new Hnsw index
    • dim (int): dimension of vectors
    • metric (std::string): an optional parameter for choosing a metric of distance. (‘L2’|‘angular’). A default metric is ‘angular’.
  • void n2::HnswIndex.AddData(const std::vector<float>& data): adds vector v
    • data (std::vector): a vector with dimension dim
  • void n2::HnswIndex.SetConfigs(const std::vector<std::pair<std::string, std::string> >& configs): Set configurations by key/value configures.
    • M (int): max number of edges for nodes at level>0 (default=12)
    • M0 (int): max number of edges for nodes at level==0 (default=24)
    • ef_construction (int): efConstruction (see HNSW paper…) (default=150)
    • n_threads (int): number of threads for building index
    • mult (float): level multiplier (recommend: use default value) (default=1/log(1.0*M))
    • neighbor_selecting (string): neighbor selecting policy
    • available values
      • NeighborSelectingPolicy::HEURISTIC(default): select neighbors using algorithm4 on HNSW paper (recommended)
      • NeighborSelectingPolicy::NAIVE: select closest neighbors (not recommended)
      • NeighborSelectingPolicy::HEURISTIC_SAVE_REMAINS: explain is needed.
    • graph_merging (string): graph merging heuristic
    • available values
      • GraphPostProcessing::SKIP (default): do not merge (recommended for large scale of data(over 10M))
      • GraphPostProcessing::MERGE_LEVEL0: build an another graph in reverse order. then merge edges of level0 (recommended for under 10M scale data)
  • void n2::HnswIndex.Fit(): builds a hnsw graph with given configurations.
  • void Build(int M = -1, int M0 = -1, int ef_construction = -1, int n_threads = -1, float mult = -1, NeighborSelectingPolicy neighbor_selecting = NeighborSelectingPolicy::HEURISTIC, GraphPostProcessing graph_merging = GraphPostProcessing::SKIP, bool ensure_k = false): builds a hnsw graph with given configurations. (see Fit, SetConfigs)
  • bool n2::HnswIndex.SaveModel(const std::string& fname): saves the index to disk
    • fname (std::string) : A index file name.
  • bool n2::HnswIndex.LoadModel(const std::string& fname, const bool use_mmap=true): loads an index from disk.
    • fname (std::string) : A index file name.
    • use_mmap(bool): An optional parameter, default value is true. If this parameter is set, N2 loads model through mmap.
  • bool n2::HnswIndex.UnloadModel(): Unload the loaded index file.
  • void n2::HnswIndex.SearchById(int id, size_t k, size_t ef_search, std::vector<std::pair<int, float> >& result): Returns k nearest items which are searched by Id.
    • ef_search (int): default value = 50 * k
  • void n2::HnswIndex.SearchByVector(const std::vector<float>& qvec, size_t k, size_t ef_search, std::vector<std::pair<int, float> >& result): Returns k nearest items which are searched by vector.
    • ef_search (int): default value = 50 * k
  • void n2::HnswIndex.PrintDegreeDist() const: Print degree distributions.
  • void n2::HnswIndex.PrintConfigs() const: Print index configurations.