Skip to content

Latest commit

 

History

History
103 lines (73 loc) · 2.53 KB

oracle.md

File metadata and controls

103 lines (73 loc) · 2.53 KB
Raw

Oracles for Common Tasks

KGraph provides the following efficient oracle implementation for common tasks. The user is encouraged to use the provided oracles when they meet the demand.

Rows of Matrices

In a very common scenario, objects are stored as rows of a matrix in the row-major order, and L2 distance is used as the similarity function. KGraph provides the class Matrix for user to create such a dataset, and the MatrixProxy class for user to directly use data managed by a OpenCV, Numpy or a FLANN matrix.

Note that the matrix proxy does not copy memory, so the original OpenCV or NumPy matrix must live throughput the life time of the matrix proxy.

OpenCV Matrix

#include <kgraph.h>
#include <opencv2/opencv.hpp>
#include <kgraph-data.h> // must follow opencv

cv::Mat data(rows, cols, CV_32FC1);

typedef kgraph::MatrixOracle<float, kgraph::metric::l2sqr
        > MyOracle;

kgraph::MatrixProxy<float> proxy(data);

MyOracle oracle(proxy);	// defined as above
index->build(oracle, ...);

// MyOracle::query(float const *) returns a search oracle
index->search(oracle.query(data.ptr<float const>(i)), ...);

NumPy Matrix

#include <Python.h>
#include <numpy/ndarrayobject.h>
#include <kgraph.h>
#include <kgraph-data.h>

npy_intp dims[] = {rows, cols};
PyArrayObject *data = PyArray_SimpleNew(2, dims, NPY_FLOAT);
kgraph::MatrixProxy<float> proxy(data);

MyOracle oracle(proxy);	// defined as above 

index->build(oracle, ...);
float const *query;
index->search(oracle.query(query), ...);

KGraph's Native Matrix

#include <kgraph.h>
#include <kgraph-data.h>

kgraph::Matrix<float> data(rows, cols);	// kgraph native matrix

data.size();   // returns # rows
data.dim();    // returns # cols
float *row = data[i];       // pointer to i-th row

// save and load matrix in LSHKIT format.
data.save_lshkit("path");
data.load_lshkit("path");

MyOracle oracle(data);	    // MyOracle defined as above

index->build(oracle, ...);

float const *query;	// pointer to query vector

// oracle.query(query) returns a search oracle.
index->search(oracle.query(query), ...)

Entries of Vectors

If data objects are stored in std::vector, then VectorOracle can come in handy.

#include <kgraph.h>

vector<MyType> data;
auto fn = [](MyType const &a, MyType const &b) {
           compute and return similarity between a and b;
          };

typedef kgraph::VectorOracle<vector<MyType>, MyType> MyOracle;
MyOracle oracle(data, fn);

index->build(oracle, ...);

MyType query;

index->search(oracle.query(query), ...);