| 1 |
/* |
| 2 |
* Copyright (c) Meta Platforms, Inc. and affiliates. |
| 3 |
* |
| 4 |
* This source code is licensed under the MIT license found in the |
| 5 |
* LICENSE file in the root directory of this source tree. |
| 6 |
*/ |
| 7 |
|
| 8 |
// -*- c -*- |
| 9 |
|
| 10 |
#ifndef FAISS_CLUSTERING_C_H |
| 11 |
#define FAISS_CLUSTERING_C_H |
| 12 |
|
| 13 |
#include "Index_c.h" |
| 14 |
#include "faiss_c.h" |
| 15 |
|
| 16 |
#ifdef __cplusplus |
| 17 |
extern "C" { |
| 18 |
#endif |
| 19 |
|
| 20 |
/** Class for the clustering parameters. Can be passed to the |
| 21 |
* constructor of the Clustering object. |
| 22 |
*/ |
| 23 |
typedef struct FaissClusteringParameters { |
| 24 |
int niter; ///< clustering iterations |
| 25 |
int nredo; ///< redo clustering this many times and keep best |
| 26 |
|
| 27 |
int verbose; ///< (bool) |
| 28 |
int spherical; ///< (bool) do we want normalized centroids? |
| 29 |
int int_centroids; ///< (bool) round centroids coordinates to integer |
| 30 |
int update_index; ///< (bool) update index after each iteration? |
| 31 |
int frozen_centroids; ///< (bool) use the centroids provided as input and do |
| 32 |
///< not change them during iterations |
| 33 |
|
| 34 |
int min_points_per_centroid; ///< otherwise you get a warning |
| 35 |
int max_points_per_centroid; ///< to limit size of dataset |
| 36 |
|
| 37 |
int seed; ///< seed for the random number generator |
| 38 |
size_t decode_block_size; ///< how many vectors at a time to decode |
| 39 |
} FaissClusteringParameters; |
| 40 |
|
| 41 |
/// Sets the ClusteringParameters object with reasonable defaults |
| 42 |
void faiss_ClusteringParameters_init(FaissClusteringParameters* params); |
| 43 |
|
| 44 |
/** clustering based on assignment - centroid update iterations |
| 45 |
* |
| 46 |
* The clustering is based on an Index object that assigns training |
| 47 |
* points to the centroids. Therefore, at each iteration the centroids |
| 48 |
* are added to the index. |
| 49 |
* |
| 50 |
* On output, the centroids table is set to the latest version |
| 51 |
* of the centroids and they are also added to the index. If the |
| 52 |
* centroids table it is not empty on input, it is also used for |
| 53 |
* initialization. |
| 54 |
* |
| 55 |
* To do several clusterings, just call train() several times on |
| 56 |
* different training sets, clearing the centroid table in between. |
| 57 |
*/ |
| 58 |
FAISS_DECLARE_CLASS(Clustering) |
| 59 |
|
| 60 |
FAISS_DECLARE_GETTER(Clustering, int, niter) |
| 61 |
FAISS_DECLARE_GETTER(Clustering, int, nredo) |
| 62 |
FAISS_DECLARE_GETTER(Clustering, int, verbose) |
| 63 |
FAISS_DECLARE_GETTER(Clustering, int, spherical) |
| 64 |
FAISS_DECLARE_GETTER(Clustering, int, int_centroids) |
| 65 |
FAISS_DECLARE_GETTER(Clustering, int, update_index) |
| 66 |
FAISS_DECLARE_GETTER(Clustering, int, frozen_centroids) |
| 67 |
|
| 68 |
FAISS_DECLARE_GETTER(Clustering, int, min_points_per_centroid) |
| 69 |
FAISS_DECLARE_GETTER(Clustering, int, max_points_per_centroid) |
| 70 |
|
| 71 |
FAISS_DECLARE_GETTER(Clustering, int, seed) |
| 72 |
FAISS_DECLARE_GETTER(Clustering, size_t, decode_block_size) |
| 73 |
|
| 74 |
/// getter for d |
| 75 |
FAISS_DECLARE_GETTER(Clustering, size_t, d) |
| 76 |
|
| 77 |
/// getter for k |
| 78 |
FAISS_DECLARE_GETTER(Clustering, size_t, k) |
| 79 |
|
| 80 |
FAISS_DECLARE_CLASS(ClusteringIterationStats) |
| 81 |
FAISS_DECLARE_GETTER(ClusteringIterationStats, float, obj) |
| 82 |
FAISS_DECLARE_GETTER(ClusteringIterationStats, double, time) |
| 83 |
FAISS_DECLARE_GETTER(ClusteringIterationStats, double, time_search) |
| 84 |
FAISS_DECLARE_GETTER(ClusteringIterationStats, double, imbalance_factor) |
| 85 |
FAISS_DECLARE_GETTER(ClusteringIterationStats, int, nsplit) |
| 86 |
|
| 87 |
/// getter for centroids (size = k * d) |
| 88 |
void faiss_Clustering_centroids( |
| 89 |
FaissClustering* clustering, |
| 90 |
float** centroids, |
| 91 |
size_t* size); |
| 92 |
|
| 93 |
/// getter for iteration stats |
| 94 |
void faiss_Clustering_iteration_stats( |
| 95 |
FaissClustering* clustering, |
| 96 |
FaissClusteringIterationStats** iteration_stats, |
| 97 |
size_t* size); |
| 98 |
|
| 99 |
/// the only mandatory parameters are k and d |
| 100 |
int faiss_Clustering_new(FaissClustering** p_clustering, int d, int k); |
| 101 |
|
| 102 |
int faiss_Clustering_new_with_params( |
| 103 |
FaissClustering** p_clustering, |
| 104 |
int d, |
| 105 |
int k, |
| 106 |
const FaissClusteringParameters* cp); |
| 107 |
|
| 108 |
int faiss_Clustering_train( |
| 109 |
FaissClustering* clustering, |
| 110 |
idx_t n, |
| 111 |
const float* x, |
| 112 |
FaissIndex* index); |
| 113 |
|
| 114 |
void faiss_Clustering_free(FaissClustering* clustering); |
| 115 |
|
| 116 |
/** simplified interface |
| 117 |
* |
| 118 |
* @param d dimension of the data |
| 119 |
* @param n nb of training vectors |
| 120 |
* @param k nb of output centroids |
| 121 |
* @param x training set (size n * d) |
| 122 |
* @param centroids output centroids (size k * d) |
| 123 |
* @param q_error final quantization error |
| 124 |
* @return error code |
| 125 |
*/ |
| 126 |
int faiss_kmeans_clustering( |
| 127 |
size_t d, |
| 128 |
size_t n, |
| 129 |
size_t k, |
| 130 |
const float* x, |
| 131 |
float* centroids, |
| 132 |
float* q_error); |
| 133 |
|
| 134 |
#ifdef __cplusplus |
| 135 |
} |
| 136 |
#endif |
| 137 |
|
| 138 |
#endif |