Misc

doc/manual.tex

@@ -279,12 +279,12 @@ must contain the following fields:
 percentage of the approximate nearest-neighbor searches that return the
 exact nearest-neighbor. Using a higher value for this parameter gives
 more
-accurate results, but the searching takes longer. The optimum value
+accurate results, but the search takes longer. The optimum value
 usually
 depends on the application.
 
 \item[\texttt{build\_weight}] - specifies the importance of the
-index build time reported to the nearest-neighbor search time. In some
+index build time raported to the nearest-neighbor search time. In some
 applications it's acceptable for the index build step to take a long time
 if the subsequent searches in the index can be performed very fast. In
 other applications it's required that the index be build as fast as
@@ -342,7 +342,7 @@ convergence. This parameter is required only when the algorithm used is
 centers when performing a kmeans clustering step. The possible values are
 'random' (picks the initial cluster centers randomly), 'gonzales' (picks the
 initial centers using the Gonzales algorithm) and 'kmeanspp' (picks the initial
-centers using thealgorithm suggested in \cite{arthur_kmeanspp_2007}). If this
+centers using the algorithm suggested in \cite{arthur_kmeanspp_2007}). If this
 parameters is omitted, the default value is 'random'.
 
 \item[\texttt{cb\_index}] - this parameter (cluster boundary index) influences the

src/cpp/CMakeLists.txt

@@ -87,7 +87,7 @@ INSTALL (
 # )
 
 INSTALL (
-    FILES flann.h constants.h
+    FILES flann.h flann.hpp constants.h util/matrix.h util/common.h util/random.h
     DESTINATION include
 )
 

src/cpp/algorithms/kdtree_index.h

@@ -34,6 +34,7 @@
 #include <algorithm>
 #include <map>
 #include <cassert>
+#include <cstring>
 
 #include "heap.h"
 #include "common.h"
@@ -175,8 +176,8 @@ public:
 	{
         size_ = dataset.rows;
         veclen_ = dataset.cols;
-
         numTrees = params.trees;
+
         trees = new Tree[numTrees];
 
 		// get the parameters
@@ -306,6 +307,7 @@ public:
      */
     void findNeighbors(ResultSet& result, const float* vec, const SearchParams& searchParams)
     {
+
         int maxChecks = searchParams.checks;
 
         if (maxChecks<0) {

src/cpp/constants.h

@@ -40,6 +40,7 @@ enum flann_algorithm_t {
 	KDTREE = 1,
 	KMEANS = 2,
 	COMPOSITE = 3,
+	KDTREE_MT = 4,
 	SAVED = 254,
 	AUTOTUNED = 255,
 };

src/cpp/flann.cpp

@@ -39,16 +39,15 @@
 #include "object_factory.h"
 // index types
 #include "kdtree_index.h"
+#include "kdtree_mt_index.h"
 #include "kmeans_index.h"
 #include "composite_index.h"
 #include "linear_index.h"
 #include "autotuned_index.h"
 
-#include <typeinfo>
 using namespace std;
 
 
-
 #include "flann.h"
 
 #ifdef WIN32
@@ -83,6 +82,11 @@ NNIndex* KDTreeIndexParams::createIndex(const Matrix<float>& dataset) const
 	return new KDTreeIndex(dataset, *this);
 }
 
+NNIndex* KDTreeMTIndexParams::createIndex(const Matrix<float>& dataset) const
+{
+	return new KDTreeMTIndex(dataset, *this);
+}
+
 NNIndex* KMeansIndexParams::createIndex(const Matrix<float>& dataset) const
 {
 	return new KMeansIndex(dataset, *this);
@@ -127,6 +131,7 @@ public:
 	{
 		ParamsFactory::instance().register_<LinearIndexParams>(LINEAR);
 		ParamsFactory::instance().register_<KDTreeIndexParams>(KDTREE);
+		ParamsFactory::instance().register_<KDTreeMTIndexParams>(KDTREE_MT);
 		ParamsFactory::instance().register_<KMeansIndexParams>(KMEANS);
 		ParamsFactory::instance().register_<CompositeIndexParams>(COMPOSITE);
 		ParamsFactory::instance().register_<AutotunedIndexParams>(AUTOTUNED);
@@ -226,7 +231,7 @@ int hierarchicalClustering(const Matrix<float>& features, Matrix<float>& centers
 
 using namespace flann;
 
-typedef NNIndex* NNIndexPtr;
+typedef Index* IndexPtr;
 typedef Matrix<float>* MatrixPtr;
 
 
@@ -259,6 +264,7 @@ EXPORTED void flann_set_distance_type(flann_distance_t distance_type, int order)
 EXPORTED flann_index_t flann_build_index(float* dataset, int rows, int cols, float* speedup, FLANNParameters* flann_params)
 {
 	try {
+
 		init_flann_parameters(flann_params);
 		if (flann_params == NULL) {
 			throw FLANNException("The flann_params argument must be non-null");
@@ -295,7 +301,7 @@ EXPORTED int flann_save_index(flann_index_t index_ptr, char* filename)
 }
 
 
-EXPORTED FLANN_INDEX flann_load_index(char* filename, float* dataset, int rows, int cols)
+EXPORTED flann_index_t flann_load_index(char* filename, float* dataset, int rows, int cols)
 {
 	try {
 		Index* index = new Index(Matrix<float>(rows,cols,dataset), SavedIndexParams(filename));
@@ -357,7 +363,7 @@ EXPORTED int flann_find_nearest_neighbors_index(flann_index_t index_ptr, float*
 }
 
 
-EXPORTED int flann_radius_search(FLANN_INDEX index_ptr,
+EXPORTED int flann_radius_search(flann_index_t index_ptr,
 										float* query,
 										int* indices,
 										float* dists,
@@ -390,7 +396,7 @@ EXPORTED int flann_radius_search(FLANN_INDEX index_ptr,
 }
 
 
-EXPORTED int flann_free_index(FLANN_INDEX index_ptr, FLANNParameters* flann_params)
+EXPORTED int flann_free_index(flann_index_t index_ptr, FLANNParameters* flann_params)
 {
 	try {
 		init_flann_parameters(flann_params);
@@ -437,7 +443,7 @@ EXPORTED void compute_ground_truth_float(float* dataset, int dshape[], float* te
 }
 
 
-EXPORTED float test_with_precision(FLANN_INDEX index_ptr, float* dataset, int dshape[], float* testset, int tshape[], int* matches, int mshape[],
+EXPORTED float test_with_precision(flann_index_t index_ptr, float* dataset, int dshape[], float* testset, int tshape[], int* matches, int mshape[],
              int nn, float precision, int* checks, int skip = 0)
 {
     assert(dshape[1]==tshape[1]);
@@ -447,8 +453,10 @@ EXPORTED float test_with_precision(FLANN_INDEX index_ptr, float* dataset, int ds
         if (index_ptr==NULL) {
             throw FLANNException("Invalid index");
         }
-        NNIndexPtr index = (NNIndexPtr)index_ptr;
-        return test_index_precision(*index, Matrix<float>(dshape[0], dshape[1],dataset), Matrix<float>(tshape[0], tshape[1], testset),
+
+        IndexPtr index = (IndexPtr)index_ptr;
+        NNIndex* nn_index = index->index();
+        return test_index_precision(*nn_index, Matrix<float>(dshape[0], dshape[1],dataset), Matrix<float>(tshape[0], tshape[1], testset),
                 Matrix<int>(mshape[0],mshape[1],matches), precision, *checks, nn, skip);
     } catch (runtime_error& e) {
         logger.error("Caught exception: %s\n",e.what());
@@ -456,7 +464,7 @@ EXPORTED float test_with_precision(FLANN_INDEX index_ptr, float* dataset, int ds
     }
 }
 
-EXPORTED float test_with_checks(FLANN_INDEX index_ptr, float* dataset, int dshape[], float* testset, int tshape[], int* matches, int mshape[],
+EXPORTED float test_with_checks(flann_index_t index_ptr, float* dataset, int dshape[], float* testset, int tshape[], int* matches, int mshape[],
              int nn, int checks, float* precision, int skip = 0)
 {
     assert(dshape[1]==tshape[1]);
@@ -466,8 +474,9 @@ EXPORTED float test_with_checks(FLANN_INDEX index_ptr, float* dataset, int dshap
         if (index_ptr==NULL) {
             throw FLANNException("Invalid index");
         }
-        NNIndexPtr index = (NNIndexPtr)index_ptr;
-        return test_index_checks(*index, Matrix<float>(dshape[0], dshape[1],dataset), Matrix<float>(tshape[0], tshape[1], testset),
+        IndexPtr index = (IndexPtr)index_ptr;
+        NNIndex* nn_index = index->index();
+        return test_index_checks(*nn_index, Matrix<float>(dshape[0], dshape[1],dataset), Matrix<float>(tshape[0], tshape[1], testset),
                 Matrix<int>(mshape[0],mshape[1],matches), checks, *precision, nn, skip);
     } catch (runtime_error& e) {
         logger.error("Caught exception: %s\n",e.what());

src/cpp/flann.h

@@ -111,7 +111,7 @@ Params:
 
 Returns: the newly created index or a number <0 for error
 */
-LIBSPEC FLANN_INDEX flann_build_index(float* dataset,
+LIBSPEC flann_index_t flann_build_index(float* dataset,
 									  int rows,
 									  int cols,
 									  float* speedup,
@@ -128,7 +128,7 @@ LIBSPEC FLANN_INDEX flann_build_index(float* dataset,
  * @param filename The filename the index should be saved to
  * @return Returns 0 on success, negative value on error.
  */
-LIBSPEC int flann_save_index(FLANN_INDEX index_id,
+LIBSPEC int flann_save_index(flann_index_t index_id,
 							 char* filename);
 
 
@@ -141,7 +141,7 @@ LIBSPEC int flann_save_index(FLANN_INDEX index_id,
  * @param cols Dataset columns
  * @return
  */
-LIBSPEC FLANN_INDEX flann_load_index(char* filename,
+LIBSPEC flann_index_t flann_load_index(char* filename,
 									 float* dataset,
 									 int rows,
 									 int cols);
@@ -192,7 +192,7 @@ Params:
 
 Returns: zero or a number <0 for error
 */
-LIBSPEC int flann_find_nearest_neighbors_index(FLANN_INDEX index_id,
+LIBSPEC int flann_find_nearest_neighbors_index(flann_index_t index_id,
 											   float* testset,
 											   int trows,
 											   int* indices,
@@ -217,7 +217,7 @@ LIBSPEC int flann_find_nearest_neighbors_index(FLANN_INDEX index_id,
  * a higher search speedup at the cost of potentially not returning all the
  * neighbours in the specified radius.
  */
-LIBSPEC int flann_radius_search(FLANN_INDEX index_ptr, /* the index */
+LIBSPEC int flann_radius_search(flann_index_t index_ptr, /* the index */
 										float* query,	/* query point */
 										int* indices, /* array for storing the indices found (will be modified) */
 										float* dists, /* similar, but for storing distances */
@@ -236,7 +236,7 @@ Params:
 
 Returns: zero or a number <0 for error
 */
-LIBSPEC int flann_free_index(FLANN_INDEX index_id,
+LIBSPEC int flann_free_index(flann_index_t index_id,
 		                     struct FLANNParameters* flann_params);
 
 /**

src/cpp/flann.hpp

@@ -59,8 +59,8 @@ public:
 
 	static IndexParams* createFromParameters(const FLANNParameters& p);
 
-	void fromParameters(const FLANNParameters& p) {};
-	void toParameters(FLANNParameters& p) { };
+	virtual void fromParameters(const FLANNParameters& p) {};
+	virtual void toParameters(FLANNParameters& p) { };
 };
 
 struct LinearIndexParams : public IndexParams {
@@ -91,6 +91,28 @@ struct KDTreeIndexParams : public IndexParams {
 
 };
 
+
+struct KDTreeMTIndexParams : public IndexParams {
+	KDTreeMTIndexParams(int trees_ = 4) : trees(trees_) {};
+
+	int trees;                 // number of randomized trees to use (for kdtree)
+
+	NNIndex* createIndex(const Matrix<float>& dataset) const;
+
+	void fromParameters(const FLANNParameters& p)
+	{
+		trees = p.trees;
+	}
+
+	void toParameters(FLANNParameters& p)
+	{
+		p.algorithm = KDTREE_MT;
+		p.trees = trees;
+	};
+
+};
+
+
 struct KMeansIndexParams : public IndexParams {
 	KMeansIndexParams(int branching_ = 32, int iterations_ = 11,
 			flann_centers_init_t centers_init_ = CENTERS_RANDOM, float cb_index_ = 0.2 ) :
@@ -236,6 +258,8 @@ public:
 	int veclen() const;
 
 	int size() const;
+
+	NNIndex* index() { return nnIndex; }
 };
 
 

src/cpp/nn/index_testing.cpp

@@ -37,6 +37,7 @@
 
 #include <algorithm>
 #include <math.h>
+#include <cstring>
 
 
 namespace flann

src/cpp/tests/CMakeLists.txt

@@ -10,7 +10,10 @@
 ADD_EXECUTABLE(flann_test flann_test.cc)
 TARGET_LINK_LIBRARIES(flann_test flann)
 
+
+
+
 INSTALL (
-    TARGETS flann_test
+    TARGETS  flann_test
     RUNTIME DESTINATION bin
 )

src/cpp/util/saving.cpp

@@ -32,6 +32,7 @@
 #include "saving.h"
 #include "nn_index.h"
 #include <cstdio>
+#include <cstring>
 
 namespace flann
 {

src/python/pyflann/bindings/flann_ctypes.py

@@ -115,7 +115,7 @@ class FLANNParameters(CustomStructure):
         'random_seed' : -1
   }
     _translation_ = {
-        "algorithm"     : {"linear"    : 0, "kdtree"    : 1, "kmeans"    : 2, "composite" : 3, "autotuned" : 255, "default"   : 1},
+        "algorithm"     : {"linear"    : 0, "kdtree"    : 1, "kmeans"    : 2, "composite" : 3, "kdtree_mt"    : 1, "autotuned" : 255, "default"   : 1},
         "centers_init"  : {"random"    : 0, "gonzales"  : 1, "kmeanspp"  : 2, "default"   : 0},
         "log_level"     : {"none"      : 0, "fatal"     : 1, "error"     : 2, "warning"   : 3, "info"      : 4, "default"   : 2}
     }
@@ -233,7 +233,7 @@ flann.compute_ground_truth_float.argtypes = [
 
 flann.test_with_precision.restype = c_float
 flann.test_with_precision.argtypes = [
-        c_void_p, 
+        FLANN_INDEX, 
         ndpointer(float32, ndim = 2, flags='aligned, c_contiguous'), # dataset
         c_int*2, # dshape
         ndpointer(float32, ndim = 2, flags='aligned, c_contiguous'), # testset
@@ -249,7 +249,7 @@ flann.test_with_precision.argtypes = [
 
 flann.test_with_checks.restype = c_float
 flann.test_with_checks.argtypes = [
-        c_void_p, 
+        FLANN_INDEX, 
         ndpointer(float32, ndim = 2, flags='aligned, c_contiguous'), # dataset
         c_int*2, # dshape
         ndpointer(float32, ndim = 2, flags='aligned, c_contiguous'), # testset
@@ -391,7 +391,7 @@ class FLANN:
         dists = empty( (nqpts, num_neighbors), dtype=float32)
                 
         self.__flann_parameters.update(kwargs)
-        
+
         flann.flann_find_nearest_neighbors(pts, npts, dim, 
             qpts, nqpts, result, dists, num_neighbors, 
             pointer(self.__flann_parameters))

src/python/pyflann/command/run_test.py

@@ -79,17 +79,23 @@ class RunTestCommand(BaseCommand):
         print 'Reading input dataset from', self.options.input_file
         dataset = read(self.options.input_file)
         
+        import time
+        print 'Creating index'
+        start_time = time.clock()
         flann = FLANN(log_level=self.options.log_level)
         flann.build_index(dataset, algorithm = self.options.algorithm,
                 trees=self.options.trees, branching=self.options.branching,
-                iterations=self.options.max_iterations, centers_init=self.options.centers_init)        
+                iterations=self.options.max_iterations, centers_init=self.options.centers_init)
+        print 'Index creation took', time.clock()-start_time         
         
         print 'Reading test dataset from', self.options.test_file
         testset = read(self.options.test_file)
         
-        print 'Reading ground truth from matches from', self.options.test_file
+        print 'Reading ground truth from matches from', self.options.match_file
         matches = read(self.options.match_file, dtype = int)
-        
+        if len(matches.shape)==1:
+            matches.shape = (matches.shape[0],1)
+
         if self.options.precision>0:
             checks, time = test_with_precision(flann, dataset, testset, matches, self.options.precision, self.options.nn)
         else:

src/python/pyflann/io/npy_dataset.py

@@ -46,7 +46,6 @@ def read(filename, dtype = numpy.float32):
         tmp = numpy.save
     except:
         raise FLANNException("Format not supported. You need at least numpy version 1.1")
-    
-    data = numpy.load(filename)    
+    data = numpy.load(filename)
     return data
     
\ No newline at end of file