/*
Creates a new R-near neighbor data structure (PRNearNeighborStructT)
from the parameters <thresholdR> and <successProbability> for the
data set <dataSet>. <nPoints> is the number of points in the data
set and <dimension> is the dimension of the points.
The set <sampleQueries> is a set with query sample points
(R-NN DS's parameters are optimized for query points from the set
<sampleQueries>). <sampleQueries> could be a sample of points from the
actual query set or from the data set. When computing the estimated
number of collisions of a sample query point <q> with the data set
points, if there is a point in the data set with the same pointer
with <q> (that is when <q> is a data set point), then the
corresponding point (<q>) is not considered in the data set (for the
purpose of computing the respective #collisions estimation).
*/
PRNearNeighborStructT initSelfTunedRNearNeighborWithDataSet(RealT thresholdR,
RealT successProbability,
Int32T nPoints,
IntT dimension,
PPointT *dataSet,
IntT nSampleQueries,
PPointT *sampleQueries,
MemVarT memoryUpperBound){
initializeLSHGlobal();
PRNearNeighborStructT nnStruct = NULL;
RNNParametersT optParameters = computeOptimalParameters(thresholdR, successProbability, nPoints, dimension, dataSet, nSampleQueries, sampleQueries, memoryUpperBound);
if (!optParameters.useUfunctions) {
DPRINTF("Used L=%d\n", optParameters.parameterL);
}else{
DPRINTF("Used m = %d\n", optParameters.parameterM);
DPRINTF("Used L = %d\n", optParameters.parameterL);
}
TimeVarT timeInit = 0;
TIMEV_START(timeInit);
// Init the R-NN data structure.
if (optParameters.typeHT != HT_HYBRID_CHAINS){
nnStruct = initLSH(optParameters, nPoints);
}else{
printRNNParameters(DEBUG_OUTPUT, optParameters);
nnStruct = initLSH_WithDataSet(optParameters, nPoints, dataSet);
}
TIMEV_END(timeInit);
DPRINTF("Time for initializing: %0.6lf\n", timeInit);
DPRINTF("Allocated memory: %lld\n", totalAllocatedMemory);
TimeVarT timeAdding = 0;
if (optParameters.typeHT != HT_HYBRID_CHAINS){
// Add the points to the LSH buckets.
TIMEV_START(timeAdding);
for(IntT i = 0; i < nPoints; i++){
addNewPointToPRNearNeighborStruct(nnStruct, dataSet[i]);
}
TIMEV_END(timeAdding);
printf("Time for adding points: %0.6lf\n", timeAdding);
DPRINTF("Allocated memory: %lld\n", totalAllocatedMemory);
}
DPRINTF("Time for creating buckets: %0.6lf\n", timeBucketCreation);
DPRINTF("Time for putting buckets into UH: %0.6lf\n", timeBucketIntoUH);
DPRINTF("Time for computing GLSH: %0.6lf\n", timeComputeULSH);
DPRINTF("NGBuckets: %d\n", nGBuckets);
return nnStruct;
}
Int32T getRNearNeighbors(PRNearNeighborStructT nnStruct, PPointT queryPoint, PPointT *(&result), Int32T &resultSize){
DPRINTF("Estimated ULSH comp: %0.6lf\n", lshPrecomp * nnStruct->nHFTuples * nnStruct->hfTuplesLength);
DPRINTF("Estimated UH overhead: %0.6lf\n", uhashOver * nnStruct->nHFTuples);
// RealT estNColls = estimateNCollisions(nnStruct->nPoints,
// nnStruct->dimension,
// nnStruct->points,
// queryPoint,
// nnStruct->parameterK,
// nnStruct->parameterL,
// nnStruct->parameterR);
// DPRINTF("Estimated #collisions (query specific): %0.6lf\n", (double)estNColls);
// estNColls = (double)estimateNDistinctCollisions(nnStruct->nPoints,
// nnStruct->dimension,
// nnStruct->points,
// queryPoint,
// nnStruct->useUfunctions,
// nnStruct->hfTuplesLength,
// nnStruct->nHFTuples,
// nnStruct->parameterR);
// DPRINTF("Estimated #distinct collisions (query specific): %0.6lf\n", estNColls);
// DPRINTF("Estimated Dist comp time (query specific): %0.6lf\n", distComp * estNColls);
// reset all the timers
timeRNNQuery = 0;
timeComputeULSH = 0;
timeGetBucket = 0;
timeCycleBucket = 0;
timeDistanceComputation = 0;
timeResultStoring = 0;
timeCycleProc = 0;
timePrecomputeHash = 0;
timeGBHash = 0;
timeChainTraversal = 0;
nOfDistComps = 0;
timeTotalBuckets = 0;
TIMEV_START(timeRNNQuery);
noExpensiveTiming = !DEBUG_PROFILE_TIMING;
Int32T nNearNeighbors = getNearNeighborsFromPRNearNeighborStruct(nnStruct, queryPoint, result, resultSize);
TIMEV_END(timeRNNQuery);
DPRINTF("Time to compute LSH: %0.6lf\n", timeComputeULSH);
DPRINTF("Time to get bucket: %0.6lf\n", timeGetBucket);
DPRINTF("Time to cycle through buckets: %0.6lf\n", timeCycleBucket);
DPRINTF("Time to for processing buckets (UH+examining points): %0.6lf\n", timeTotalBuckets);
//DPRINTF("Time to copy ULSHs: %0.6lf\n", timeCopyingULSHs);
//DPRINTF("Time to unmark points: %0.6lf\n", timeUnmarking);
DPRINTF("Time for distance comps: %0.6lf\n", timeDistanceComputation);
DPRINTF("Time to store result: %0.6lf\n", timeResultStoring);
//printf("Time for cycle processing: %0.6lf\n", timeCycleProc);
//printf("Time for precomputing hashes: %0.6lf\n", timePrecomputeHash);
//printf("Time for GB hash: %0.6lf\n", timeGBHash);
//printf("Time for traversal of chains: %0.6lf\n", timeChainTraversal);
DPRINTF("Number of dist comps: %d\n", nOfDistComps);
DPRINTF("Number buckets in chains: %d\n", nBucketsInChains);
DPRINTF("Number buckets in chains / L: %0.3lf\n", (double)nBucketsInChains / nnStruct->nHFTuples);
DPRINTF("Cumulative time for R-NN query: %0.6lf\n", timeRNNQuery);
return nNearNeighbors;
}
// Returns TRUE iff |p1-p2|_2^2 <= threshold
inline BooleanT isDistanceSqrLeq(IntT dimension, PPointT p1, PPointT p2, RealT threshold){
RealT result = 0;
nOfDistComps++;
TIMEV_START(timeDistanceComputation);
for (IntT i = 0; i < dimension; i++){
RealT temp = p1->coordinates[i] - p2->coordinates[i];
result += SQR(temp);
if (result > threshold){
TIMEV_END(timeDistanceComputation);
return 0;
}
}
TIMEV_END(timeDistanceComputation);
//return result <= threshold;
return 1;
}
// Computes how much time it takes to run timing functions (functions
// that compute timings) -- we need to substract this value when we
// compute the length of an actual interval of time.
//计算函数运行的时间，我们需要减去这个值，得到精确的处理时间
void tuneTimeFunctions(){
timevSpeed = 0;
// Compute the time needed for a calls to TIMEV_START and TIMEV_END
IntT nIterations = 100000;
TimeVarT timeVar = 0;
for(IntT i = 0; i < nIterations; i++){
TIMEV_START(timeVar);
TIMEV_END(timeVar);
}
timevSpeed = timeVar / nIterations;
DPRINTF("Tuning: timevSpeed = %0.9lf\n", timevSpeed);
}
int main(int nargs, char **args){
if (nargs < 7) {
usage(args[0]);
exit(1);
}
nPoints = atoi(args[1]);
nQueries = atoi(args[2]);
dimension = atoi(args[3]);
p = atof(args[4]);
K = atoi(args[5]);
readPoints(args[6]); // read all points
FILE *queryFile = fopen(args[7], "rt");
//fscanf(queryFile, "%d\n", &nQueries);
query = (RealT*)malloc(dimension * sizeof(RealT));
printf("nPoints = %d\n", nPoints);
//printf("nQueries = %d\n", nQueries);
for(int i = 0; i < nQueries; i++){
// read in the query point.
for(int d = 0; d < dimension; d++){
FSCANF_REAL(queryFile, &(query[d]));
}
//printRealVector1("Query: ", dimension, query);
std::priority_queue<Node> myq;
TimeVarT time = 0;
RealT tempdis = 0;
TIMEV_START(time);
for(int j = 0; j < nPoints; j++){
tempdis = dist(query, points[j]);
updataQ(myq, tempdis, j);
//printf("Distance[dist] (%d): %lf\n", j, dist(query, points[j]));
//printRealVector1("X: ", dimension, points[j]);
}
TIMEV_END(time); // time only finding the near neighbors, and exclude printing from timing.
printf("Total time for K-NN query \t%0.6lf\n",time);
printf("Query point %d 's %d NNs are:\n", i, K);
display(myq);
}
}
// Adds a new point to the LSH data structure, that is for each
// i=0..parameterL-1, the point is added to the bucket defined by
// function g_i=lshFunctions[i].
void addNewPointToPRNearNeighborStruct(PRNearNeighborStructT nnStruct, PPointT point){
ASSERT(nnStruct != NULL);
ASSERT(point != NULL);
ASSERT(nnStruct->reducedPoint != NULL);
ASSERT(!nnStruct->useUfunctions || nnStruct->pointULSHVectors != NULL);
ASSERT(nnStruct->hashedBuckets[0]->typeHT == HT_LINKED_LIST || nnStruct->hashedBuckets[0]->typeHT == HT_STATISTICS);
nnStruct->points[nnStruct->nPoints] = point;
nnStruct->nPoints++;
preparePointAdding(nnStruct, nnStruct->hashedBuckets[0], point);
// Initialize the counters for defining the pair of <u> functions used for <g> functions.
IntT firstUComp = 0;
IntT secondUComp = 1;
TIMEV_START(timeBucketIntoUH);
for(IntT i = 0; i < nnStruct->parameterL; i++){
if (!nnStruct->useUfunctions) {
// Use usual <g> functions (truly independent; <g>s are precisly
// <u>s).
addBucketEntry(nnStruct->hashedBuckets[i], 1, nnStruct->precomputedHashesOfULSHs[i], NULL, nnStruct->nPoints - 1);
} else {
// Use <u> functions (<g>s are pairs of <u> functions).
addBucketEntry(nnStruct->hashedBuckets[i], 2, nnStruct->precomputedHashesOfULSHs[firstUComp], nnStruct->precomputedHashesOfULSHs[secondUComp], nnStruct->nPoints - 1);
// compute what is the next pair of <u> functions.
secondUComp++;
if (secondUComp == nnStruct->nHFTuples) {
firstUComp++;
secondUComp = firstUComp + 1;
}
}
//batchAddRequest(nnStruct, i, firstUComp, secondUComp, point);
}
TIMEV_END(timeBucketIntoUH);
// Check whether the vectors <nearPoints> & <nearPointsIndeces> is still big enough.
if (nnStruct->nPoints > nnStruct->sizeMarkedPoints) {
nnStruct->sizeMarkedPoints = 2 * nnStruct->nPoints;
FAILIF(NULL == (nnStruct->markedPoints = (BooleanT*)REALLOC(nnStruct->markedPoints, nnStruct->sizeMarkedPoints * sizeof(BooleanT))));
for(IntT i = 0; i < nnStruct->sizeMarkedPoints; i++){
nnStruct->markedPoints[i] = FALSE;
}
FAILIF(NULL == (nnStruct->markedPointsIndeces = (Int32T*)REALLOC(nnStruct->markedPointsIndeces, nnStruct->sizeMarkedPoints * sizeof(Int32T))));
}
}
inline void preparePointAdding(PRNearNeighborStructT nnStruct, PUHashStructureT uhash, PPointT point)
{
//输入： nnntstuct结构体（降维的向量）， uhash（hash的两个主副向量） 特征点
//操作：先计算点的降维结果，然后计算两个hash索引值 保存到nnStruct->precomputedHashesOfULSHs
//功能，提前计算好每个点的hash表索引值
//根据传入的多维point。
//计算对应每个hash表的降维=》hash值，
//存入了nnStruct->precomputedHashesOfULSHs
ASSERT(nnStruct != NULL);
ASSERT(uhash != NULL);
ASSERT(point != NULL);
TIMEV_START(timeComputeULSH);
for(IntT d = 0; d < nnStruct->dimension; d++) {
nnStruct->reducedPoint[d] = point->coordinates[d] / nnStruct->parameterR;
}
//降维
// Compute all ULSH functions.
for(IntT i = 0; i < nnStruct->nHFTuples; i++) { //nHFTuples是元组个数，
//求出nnStruct->reducedPoint向量和多个hansh映射后的值， 对于每个hash： a。v+b 除以 r
//结果返回到pointULSHVectors【】 向量上
//pointULSHVectors【i】就是 第一步降维后的向量值
computeULSH(nnStruct, i, nnStruct->reducedPoint, nnStruct->pointULSHVectors[i]);
}
//模hash
// Compute data for <precomputedHashesOfULSHs>.
if (USE_SAME_UHASH_FUNCTIONS) {
for(IntT i = 0; i < nnStruct->nHFTuples; i++) {
precomputeUHFsForULSH(uhash, nnStruct->pointULSHVectors[i], nnStruct->hfTuplesLength, nnStruct->precomputedHashesOfULSHs[i]);
//根据降维后的结果向量：pointULSHVectors【i】 计算两个hash值，存入precomputedHashesOfULSHs【i】
}
}
TIMEV_END(timeComputeULSH);
}
// Returns TRUE iff |p1-p2|_2^2 <= threshold
inline BooleanT isDistanceSqrLeq(IntT dimension, PPointT p1, PPointT p2, RealT threshold)
{ //直接计算：|p1-p2|_2^2 <= threshold；两点距离是否小于阈值
RealT result = 0;
nOfDistComps++;
TIMEV_START(timeDistanceComputation);
for (IntT i = 0; i < dimension; i++) {
RealT temp = p1->coordinates[i] - p2->coordinates[i];
#ifdef USE_L1_DISTANCE
result += ABS(temp);
#else
result += SQR(temp);
#endif
if (result > threshold) {
// TIMEV_END(timeDistanceComputation);
return 0;
}
}
TIMEV_END(timeDistanceComputation);
//return result <= threshold;
return 1;
}
inline void preparePointAdding(PRNearNeighborStructT nnStruct, PUHashStructureT uhash, PPointT point){
ASSERT(nnStruct != NULL);
ASSERT(uhash != NULL);
ASSERT(point != NULL);
TIMEV_START(timeComputeULSH);
for(IntT d = 0; d < nnStruct->dimension; d++){
nnStruct->reducedPoint[d] = point->coordinates[d] / nnStruct->parameterR;
}
// Compute all ULSH functions.
for(IntT i = 0; i < nnStruct->nHFTuples; i++){
computeULSH(nnStruct, i, nnStruct->reducedPoint, nnStruct->pointULSHVectors[i]);
}
// Compute data for <precomputedHashesOfULSHs>.
if (USE_SAME_UHASH_FUNCTIONS) {
for(IntT i = 0; i < nnStruct->nHFTuples; i++){
precomputeUHFsForULSH(uhash, nnStruct->pointULSHVectors[i], nnStruct->hfTuplesLength, nnStruct->precomputedHashesOfULSHs[i]);
}
}
TIMEV_END(timeComputeULSH);
}
// Returns the list of near neighbors of the point <point> (with a
// certain success probability). Near neighbor is defined as being a
// point within distance <parameterR>. Each near neighbor from the
// data set is returned is returned with a certain probability,
// dependent on <parameterK>, <parameterL>, and <parameterT>. The
// returned points are kept in the array <result>. If result is not
// allocated, it will be allocated to at least some minimum size
// (RESULT_INIT_SIZE). If number of returned points is bigger than the
// size of <result>, then the <result> is resized (to up to twice the
// number of returned points). The return value is the number of
// points found.
Int32T getNearNeighborsFromPRNearNeighborStruct(PRNearNeighborStructT nnStruct, PPointT query, PPointT *(&result), Int32T &resultSize){
ASSERT(nnStruct != NULL);
ASSERT(query != NULL);
ASSERT(nnStruct->reducedPoint != NULL);
ASSERT(!nnStruct->useUfunctions || nnStruct->pointULSHVectors != NULL);
PPointT point = query;
if (result == NULL){
resultSize = RESULT_INIT_SIZE;
FAILIF(NULL == (result = (PPointT*)MALLOC(resultSize * sizeof(PPointT))));
}
preparePointAdding(nnStruct, nnStruct->hashedBuckets[0], point);
Uns32T precomputedHashesOfULSHs[nnStruct->nHFTuples][N_PRECOMPUTED_HASHES_NEEDED];
for(IntT i = 0; i < nnStruct->nHFTuples; i++){
for(IntT j = 0; j < N_PRECOMPUTED_HASHES_NEEDED; j++){
precomputedHashesOfULSHs[i][j] = nnStruct->precomputedHashesOfULSHs[i][j];
}
}
TIMEV_START(timeTotalBuckets);
BooleanT oldTimingOn = timingOn;
if (noExpensiveTiming) {
timingOn = FALSE;
}
// Initialize the counters for defining the pair of <u> functions used for <g> functions.
IntT firstUComp = 0;
IntT secondUComp = 1;
Int32T nNeighbors = 0;// the number of near neighbors found so far.
Int32T nMarkedPoints = 0;// the number of marked points
for(IntT i = 0; i < nnStruct->parameterL; i++){
TIMEV_START(timeGetBucket);
GeneralizedPGBucket gbucket;
if (!nnStruct->useUfunctions) {
// Use usual <g> functions (truly independent; <g>s are precisly
// <u>s).
gbucket = getGBucket(nnStruct->hashedBuckets[i], 1, precomputedHashesOfULSHs[i], NULL);
} else {
// Use <u> functions (<g>s are pairs of <u> functions).
gbucket = getGBucket(nnStruct->hashedBuckets[i], 2, precomputedHashesOfULSHs[firstUComp], precomputedHashesOfULSHs[secondUComp]);
// compute what is the next pair of <u> functions.
secondUComp++;
if (secondUComp == nnStruct->nHFTuples) {
firstUComp++;
secondUComp = firstUComp + 1;
}
}
TIMEV_END(timeGetBucket);
PGBucketT bucket;
TIMEV_START(timeCycleBucket);
switch (nnStruct->hashedBuckets[i]->typeHT){
case HT_LINKED_LIST:
bucket = gbucket.llGBucket;
if (bucket != NULL){
// circle through the bucket and add to <result> the points that are near.
PBucketEntryT bucketEntry = &(bucket->firstEntry);
//TIMEV_START(timeCycleProc);
while (bucketEntry != NULL){
//TIMEV_END(timeCycleProc);
//ASSERT(bucketEntry->point != NULL);
//TIMEV_START(timeDistanceComputation);
Int32T candidatePIndex = bucketEntry->pointIndex;
PPointT candidatePoint = nnStruct->points[candidatePIndex];
if (isDistanceSqrLeq(nnStruct->dimension, point, candidatePoint, nnStruct->parameterR2) && nnStruct->reportingResult){
//TIMEV_END(timeDistanceComputation);
if (nnStruct->markedPoints[candidatePIndex] == FALSE) {
//TIMEV_START(timeResultStoring);
// a new R-NN point was found (not yet in <result>).
if (nNeighbors >= resultSize){
// run out of space => resize the <result> array.
resultSize = 2 * resultSize;
result = (PPointT*)REALLOC(result, resultSize * sizeof(PPointT));
}
result[nNeighbors] = candidatePoint;
nNeighbors++;
nnStruct->markedPointsIndeces[nMarkedPoints] = candidatePIndex;
nnStruct->markedPoints[candidatePIndex] = TRUE; // do not include more points with the same index
nMarkedPoints++;
//TIMEV_END(timeResultStoring);
}
}else{
//TIMEV_END(timeDistanceComputation);
}
//TIMEV_START(timeCycleProc);
bucketEntry = bucketEntry->nextEntry;
}
//TIMEV_END(timeCycleProc);
}
break;
case HT_STATISTICS:
ASSERT(FALSE); // HT_STATISTICS not supported anymore
// if (gbucket.linkGBucket != NULL && gbucket.linkGBucket->indexStart != INDEX_START_EMPTY){
// Int32T position;
// PointsListEntryT *pointsList = nnStruct->hashedBuckets[i]->bucketPoints.pointsList;
// position = gbucket.linkGBucket->indexStart;
// // circle through the bucket and add to <result> the points that are near.
// while (position != INDEX_START_EMPTY){
// PPointT candidatePoint = pointsList[position].point;
// if (isDistanceSqrLeq(nnStruct->dimension, point, candidatePoint, nnStruct->parameterR2) && nnStruct->reportingResult){
// if (nnStruct->nearPoints[candidatePoint->index] == FALSE) {
// // a new R-NN point was found (not yet in <result>).
// if (nNeighbors >= resultSize){
// // run out of space => resize the <result> array.
// resultSize = 2 * resultSize;
// result = (PPointT*)REALLOC(result, resultSize * sizeof(PPointT));
// }
// result[nNeighbors] = candidatePoint;
// nNeighbors++;
// nnStruct->nearPoints[candidatePoint->index] = TRUE; // do not include more points with the same index
// }
// }
// // Int32T oldP = position;
// position = pointsList[position].nextPoint;
// // ASSERT(position == INDEX_START_EMPTY || position == oldP + 1);
// }
// }
break;
case HT_HYBRID_CHAINS:
if (gbucket.hybridGBucket != NULL){
PHybridChainEntryT hybridPoint = gbucket.hybridGBucket;
Uns32T offset = 0;
if (hybridPoint->point.bucketLength == 0){
// there are overflow points in this bucket.
offset = 0;
for(IntT j = 0; j < N_FIELDS_PER_INDEX_OF_OVERFLOW; j++){
offset += ((Uns32T)((hybridPoint + 1 + j)->point.bucketLength) << (j * N_BITS_FOR_BUCKET_LENGTH));
}
}
Uns32T index = 0;
BooleanT done = FALSE;
while(!done){
if (index == MAX_NONOVERFLOW_POINTS_PER_BUCKET){
//CR_ASSERT(hybridPoint->point.bucketLength == 0);
index = index + offset;
}
Int32T candidatePIndex = (hybridPoint + index)->point.pointIndex;
CR_ASSERT(candidatePIndex >= 0 && candidatePIndex < nnStruct->nPoints);
done = (hybridPoint + index)->point.isLastPoint == 1 ? TRUE : FALSE;
index++;
if (nnStruct->markedPoints[candidatePIndex] == FALSE){
// mark the point first.
nnStruct->markedPointsIndeces[nMarkedPoints] = candidatePIndex;
nnStruct->markedPoints[candidatePIndex] = TRUE; // do not include more points with the same index
nMarkedPoints++;
PPointT candidatePoint = nnStruct->points[candidatePIndex];
if (isDistanceSqrLeq(nnStruct->dimension, point, candidatePoint, nnStruct->parameterR2) && nnStruct->reportingResult){
//if (nnStruct->markedPoints[candidatePIndex] == FALSE) {
// a new R-NN point was found (not yet in <result>).
//TIMEV_START(timeResultStoring);
if (nNeighbors >= resultSize){
// run out of space => resize the <result> array.
resultSize = 2 * resultSize;
result = (PPointT*)REALLOC(result, resultSize * sizeof(PPointT));
}
result[nNeighbors] = candidatePoint;
nNeighbors++;
//TIMEV_END(timeResultStoring);
//nnStruct->markedPointsIndeces[nMarkedPoints] = candidatePIndex;
//nnStruct->markedPoints[candidatePIndex] = TRUE; // do not include more points with the same index
//nMarkedPoints++;
//}
}
}else{
// the point was already marked (& examined)
}
}
}
break;
default:
ASSERT(FALSE);
}
TIMEV_END(timeCycleBucket);
}
timingOn = oldTimingOn;
TIMEV_END(timeTotalBuckets);
// we need to clear the array nnStruct->nearPoints for the next query.
for(Int32T i = 0; i < nMarkedPoints; i++){
ASSERT(nnStruct->markedPoints[nnStruct->markedPointsIndeces[i]] == TRUE);
nnStruct->markedPoints[nnStruct->markedPointsIndeces[i]] = FALSE;
}
DPRINTF("nMarkedPoints: %d\n", nMarkedPoints);
return nNeighbors;
}
// Returns the list of near neighbors of the point <point> (with a
// certain success probability). Near neighbor is defined as being a
// point within distance <parameterR>. Each near neighbor from the
// data set is returned is returned with a certain probability,
// dependent on <parameterK>, <parameterL>, and <parameterT>. The
// returned points are kept in the array <result>. If result is not
// allocated, it will be allocated to at least some minimum size
// (RESULT_INIT_SIZE). If number of returned points is bigger than the
// size of <result>, then the <result> is resized (to up to twice the
// number of returned points). The return value is the number of
// points found.
Int32T getNearNeighborsFromPRNearNeighborStruct(
PRNearNeighborStructT nnStruct, PPointT query,
PPointT *(&result), Int32T &resultSize)
{ //通过查找索引，然后获得桶，提取n个最近邻点
//通过计算点的降维值，然后计算主副索引，最后由索引查找表
ASSERT(nnStruct != NULL);
ASSERT(query != NULL);
ASSERT(nnStruct->reducedPoint != NULL);
ASSERT(!nnStruct->useUfunctions || nnStruct->pointULSHVectors != NULL);
PPointT point = query;
if (result == NULL)
{
resultSize = RESULT_INIT_SIZE;
FAILIF(NULL == (result = (PPointT*)MALLOC(resultSize * sizeof(PPointT))));
}
/*
for (int tempd=150; tempd< 160;tempd++)
{
printf(" %lf ",query->coordinates[tempd]);
}
printf("查询的具体数据 10个 \n\n");
printf("查询数据 : %lf \n",query->coordinates[151]);
// printf( "主hash的值： %u \n",nnStruct->hehasdBuckets[0]->mainHashA[5]);
// printf( "辅助hash的值： %u \n",nnStruct->hashedBuckets[0]->controlHash1[5]);
// printf( "a %u \n",nnStruct->lshFunctions[0][0].a[5]);
// printf( "b %u \n",nnStruct->lshFunctions[0][0].b );
*/
preparePointAdding(nnStruct, nnStruct->hashedBuckets[0], point);
//根据传入的多维point。计算对应每个hash表的降维=》hash值，存入了nnStruct->precomputedHashesOfULSHs
Uns32T **(precomputedHashesOfULSHs);//没释放
precomputedHashesOfULSHs= (Uns32T**)malloc(sizeof(Uns32T*)*(nnStruct->nHFTuples));
// Uns32T precomputedHashesOfULSHs[nnStruct->nHFTuples][N_PRECOMPUTED_HASHES_NEEDED];
for (IntT i=0; i< nnStruct->nHFTuples ; i++)
{
precomputedHashesOfULSHs[i]= (Uns32T*)malloc(sizeof(Uns32T)*(N_PRECOMPUTED_HASHES_NEEDED));
for (int temi=0; temi< N_PRECOMPUTED_HASHES_NEEDED ; temi++)
{
precomputedHashesOfULSHs[i][temi]=0;
}
}
//初始化？？
/*
printf("\n输出：\n");
FILE *in = fopen("preconpute.txt", "a+") ;
fprintf(in,"\n输出：\n");
fclose(in);
*/
for(IntT i = 0; i < nnStruct->nHFTuples; i++)
{
for(IntT j = 0; j < N_PRECOMPUTED_HASHES_NEEDED; j++)
{
precomputedHashesOfULSHs[i][j] = nnStruct->precomputedHashesOfULSHs[i][j];
/* printf(" %u", precomputedHashesOfULSHs[i][j]);
FILE *in = fopen("preconpute.txt", "a+") ;
fprintf(in," %u", precomputedHashesOfULSHs[i][j]);
fclose(in);
*/
}
/*printf(" \n");
FILE *in = fopen("preconpute.txt", "a+") ;
fprintf(in," \n");
fclose(in);
*/
}
TIMEV_START(timeTotalBuckets);
BooleanT oldTimingOn = timingOn;
if (noExpensiveTiming)
{
timingOn = FALSE;
}
// Initialize the counters for defining the pair of <u> functions used for <g> functions.
IntT firstUComp = 0;
IntT secondUComp = 1;
Int32T nNeighbors = 0;// the number of near neighbors found so far.
Int32T nMarkedPoints = 0;// the number of marked points
for(IntT i = 0; i < nnStruct->parameterL; i++)
{ //L个表
TIMEV_START(timeGetBucket);
GeneralizedPGBucket gbucket;
if (!nnStruct->useUfunctions)
{
// Use usual <g> functions (truly independent; <g>s are precisly
// <u>s).
gbucket = getGBucket(nnStruct->hashedBuckets[i], 1, precomputedHashesOfULSHs[i], NULL);
}
else
{
// Use <u> functions (<g>s are pairs of <u> functions).
gbucket = getGBucket(nnStruct->hashedBuckets[i], 2, precomputedHashesOfULSHs[firstUComp], precomputedHashesOfULSHs[secondUComp]);
//通过两个向量，计算主副索引。然后遍历二级索引，提取对应的桶
// compute what is the next pair of <u> functions.
//不是每个都 （first，second ）（first，second ）（first，second ）的数组吗？
secondUComp++;
if (secondUComp == nnStruct->nHFTuples)
{
firstUComp++;
secondUComp = firstUComp + 1;
}
}
TIMEV_END(timeGetBucket);
PGBucketT bucket;
TIMEV_START(timeCycleBucket);
switch (nnStruct->hashedBuckets[i]->typeHT)
{ //对不同类型的hash桶结构，使用不同方法获取二级桶的实体
case HT_LINKED_LIST:
bucket = gbucket.llGBucket;
if (bucket != NULL)
{
// circle through the bucket and add to <result> the points that are near.
PBucketEntryT bucketEntry = &(bucket->firstEntry);
//TIMEV_START(timeCycleProc);
while (bucketEntry != NULL)
{
//TIMEV_END(timeCycleProc);
//ASSERT(bucketEntry->point != NULL);
//TIMEV_START(timeDistanceComputation);
Int32T candidatePIndex = bucketEntry->pointIndex;
PPointT candidatePoint = nnStruct->points[candidatePIndex];
if (isDistanceSqrLeq(nnStruct->dimension, point, candidatePoint, nnStruct->parameterR2)
&& nnStruct->reportingResult)
{
//TIMEV_END(timeDistanceComputation);
if (nnStruct->markedPoints[candidatePIndex] == FALSE)
{
//TIMEV_START(timeResultStoring);
// a new R-NN point was found (not yet in <result>).
if (nNeighbors >= resultSize)
{
// run out of space => resize the <result> array.
resultSize = 2 * resultSize;
result = (PPointT*)REALLOC(result, resultSize * sizeof(PPointT));
}
result[nNeighbors] = candidatePoint;
nNeighbors++;
nnStruct->markedPointsIndeces[nMarkedPoints] = candidatePIndex;
nnStruct->markedPoints[candidatePIndex] = TRUE; // do not include more points with the same index
nMarkedPoints++;
//TIMEV_END(timeResultStoring);
}
}
else
{
//TIMEV_END(timeDistanceComputation);
}
//TIMEV_START(timeCycleProc);
bucketEntry = bucketEntry->nextEntry;
}//while
//TIMEV_END(timeCycleProc);
}
break;
case HT_STATISTICS:
ASSERT(FALSE); // HT_STATISTICS not supported anymore
// if (gbucket.linkGBucket != NULL && gbucket.linkGBucket->indexStart != INDEX_START_EMPTY){
// Int32T position;
// PointsListEntryT *pointsList = nnStruct->hashedBuckets[i]->bucketPoints.pointsList;
// position = gbucket.linkGBucket->indexStart;
// // circle through the bucket and add to <result> the points that are near.
// while (position != INDEX_START_EMPTY){
// PPointT candidatePoint = pointsList[position].point;
// if (isDistanceSqrLeq(nnStruct->dimension, point, candidatePoint, nnStruct->parameterR2) && nnStruct->reportingResult){
// if (nnStruct->nearPoints[candidatePoint->index] == FALSE) {
// // a new R-NN point was found (not yet in <result>).
// if (nNeighbors >= resultSize){
// // run out of space => resize the <result> array.
// resultSize = 2 * resultSize;
// result = (PPointT*)REALLOC(result, resultSize * sizeof(PPointT));
// }
// result[nNeighbors] = candidatePoint;
// nNeighbors++;
// nnStruct->nearPoints[candidatePoint->index] = TRUE; // do not include more points with the same index
// }
// }
// // Int32T oldP = position;
// position = pointsList[position].nextPoint;
// // ASSERT(position == INDEX_START_EMPTY || position == oldP + 1);
// }
// }
break;
case HT_HYBRID_CHAINS://默认的链条
if (gbucket.hybridGBucket != NULL)
{ //好像是在链表中找空间，同时要判断没有重复的
PHybridChainEntryT hybridPoint = gbucket.hybridGBucket;//获取 二级桶的数组指针，（实际桶就是一个数组）
Uns32T offset = 0;
if (hybridPoint->point.bucketLength == 0)
{ //长度为0，就是溢出了的桶，
// there are overflow points in this bucket.
offset = 0;
for(IntT j = 0; j < N_FIELDS_PER_INDEX_OF_OVERFLOW; j++)
{
offset += ((Uns32T)((hybridPoint + 1 + j)->point.bucketLength) << (j * N_BITS_FOR_BUCKET_LENGTH));
}
}
Uns32T index = 0;
BooleanT done = FALSE;
while(!done)
{
if (index == MAX_NONOVERFLOW_POINTS_PER_BUCKET)
{
//CR_ASSERT(hybridPoint->point.bucketLength == 0);
index = index + offset;
}
//hybridPoint 是个二级桶+实体组成的数组的首地址（其实就是个二级刻度）
Int32T candidatePIndex = (hybridPoint + index)->point.pointIndex;
//索引只是记录每个点的序号， 所有点都在nnStruct->points[candidatePIndex] 上保存具体值
CR_ASSERT(candidatePIndex >= 0 && candidatePIndex < nnStruct->nPoints);
done = (hybridPoint + index)->point.isLastPoint == 1 ? TRUE : FALSE;
//链表的遍历？好像是用数组来当链表用
index++;
if (nnStruct->markedPoints[candidatePIndex] == FALSE)
{ //已经计算过的点都标记为true了
//nnStruct->markedPoints 是用来标记是否检测过得
// mark the point first.
nnStruct->markedPointsIndeces[nMarkedPoints] = candidatePIndex;
nnStruct->markedPoints[candidatePIndex] = TRUE; // do not include more points with the same index
nMarkedPoints++;
PPointT candidatePoint = nnStruct->points[candidatePIndex];
if (isDistanceSqrLeq(nnStruct->dimension, point, candidatePoint, nnStruct->parameterR2)
&& nnStruct->reportingResult)
{ //两点距离是否小于阈值
//if (nnStruct->markedPoints[candidatePIndex] == FALSE) {
// a new R-NN point was found (not yet in <result>).
//TIMEV_START(timeResultStoring);
if (nNeighbors >= resultSize)
{ //近邻点太多，扩大空间
// run out of space => resize the <result> array.
resultSize = 2 * resultSize;
result = (PPointT*)REALLOC(result, resultSize * sizeof(PPointT));
}
result[nNeighbors] = candidatePoint;//存入返回结果中
nNeighbors++;
//TIMEV_END(timeResultStoring);
//nnStruct->markedPointsIndeces[nMarkedPoints] = candidatePIndex;
//nnStruct->markedPoints[candidatePIndex] = TRUE; // do not include more points with the same index
//nMarkedPoints++;
//}
}
}// if (nnStruct->markedPoints[candidatePIndex] == FALSE)
else
{
// the point was already marked (& examined)
}
}// while(!done)
}// if (gbucket.hybridGBucket != NULL)
break;
default:
ASSERT(FALSE);
}//swichcase
TIMEV_END(timeCycleBucket);
}//for
timingOn = oldTimingOn;
TIMEV_END(timeTotalBuckets);
// we need to clear the array nnStruct->nearPoints for the next query.
for(Int32T i = 0; i < nMarkedPoints; i++)
{
ASSERT(nnStruct->markedPoints[nnStruct->markedPointsIndeces[i]] == TRUE);
nnStruct->markedPoints[nnStruct->markedPointsIndeces[i]] = FALSE;
}
DPRINTF("nMarkedPoints: %d\n", nMarkedPoints);
return nNeighbors;
}