double closestPointsf(vector<R3> & src, vector<R3> & dst, vector<int> & indexes, vector<double> & distances)
{
distances.clear();
indexes.clear();
#ifndef USE_KD
Mat srcm = asMatRows(src), dstm = asMatRows(dst);
vector<float> dists;
double error = (double) knn(dstm, srcm, indexes, dists);
for(int i = 0; i < dists.size(); i++) distances.push_back((double)dists[i]);
double divisor = src.size() > 0 ? (double)(src.size()) : 1.0;
return error / divisor;
#else
Pixel3DSet p1(src);
Pixel3DSet p2(dst);
LKDNode n; n.init(p2);
n.split(p2);
double error = 0.0;
for(int i = 0; i < src.size(); i++)
{
R3 p = src[i];
R3 w;
int index = 0;
n.NN(p2,p, index, w);
indexes.push_back(index);
float D = p.dist(p2[index]);
distances.push_back(D);
error += (D*D) / (float) src.size();
}
return error;
#endif
}
void TEST_knn(char *file, int size_database, int quantity_atributes, int k) {
int size_class_values = 2;
struct database *data = mount_database(file, size_database, quantity_atributes);
assert(data);
data->class_values = (double *)calloc(size_class_values, sizeof(double));
assert(data->class_values);
data->size_class_values = size_class_values;
data->class_values[0] = 1.0;
data->class_values[1] = 0.0;
int **confusion_matrix = create_confusion_matrix();
knn(data, k, confusion_matrix);
printf("TP:%d, FN:%d\nFP:%d, TN:%d\n",
confusion_matrix[0][0], confusion_matrix[0][1],
confusion_matrix[1][0], confusion_matrix[1][1]);
double i = (double)(confusion_matrix[0][0] + confusion_matrix[1][1]);
double j = (double)(confusion_matrix[0][0] + confusion_matrix[0][1] +
confusion_matrix[1][0] + confusion_matrix[1][1]);
printf("k:%d\naccuracy: %f\n", k, (double)(i/j));
free_database(data);
free_confusion_matrix(confusion_matrix);
}
static int
knn_class(flt *x, struct codebook *cb, int nclass, int k)
{
int *tbl = 0;
int *ctab = 0;
int i,n,c;
tbl=(int*)malloc(k*sizeof(int));
ctab=(int*)malloc(nclass*sizeof(int));
ifn ( tbl && ctab) {
if (tbl) free(tbl);
if (ctab) free(ctab);
error(NIL,"No memory",NIL);
}
knn(x, cb, k, tbl);
for (i=0; i<nclass; i++)
ctab[i] = 0;
for (i=0; i<k; i++) {
c = cb->code[tbl[i]].label;
if (c>=0 && c<nclass)
ctab[c]++;
}
n = c = -1;
for (i=0; i<nclass; i++)
if (ctab[i]>n) {
c = i;
n = ctab[i];
} else if (ctab[i]==n) {
c = -1;
}
if (tbl) free(tbl);
if (ctab) free(ctab);
return c;
}
void MLearn::runML(){
Mat trainingData ( numTrainingPoints , 2 , CV_32FC1 ) ;
Mat testData ( numTestPoints , 2 , CV_32FC1 ) ;
randu ( trainingData ,0 ,1) ;
randu ( testData ,0 ,1) ;
Mat trainingClasses = labelData ( trainingData , eq ) ;
Mat testClasses = labelData ( testData , eq ) ;
plot_binary ( trainingData , trainingClasses , " Training Data " ) ;
plot_binary ( testData , testClasses , " Test Data " ) ;
svm ( trainingData , trainingClasses , testData , testClasses ) ;
mlp ( trainingData , trainingClasses , testData , testClasses ) ;
knn ( trainingData , trainingClasses , testData , testClasses , 3) ;
bayes ( trainingData , trainingClasses , testData , testClasses ) ;
decisiontree ( trainingData , trainingClasses , testData , testClasses ) ;
waitKey () ;
}
double closestPointsf(Pixel3DSet & src, Pixel3DSet & dst, vector<int> & indexes, vector<double> & distances)
{
//cout << "timing " << endl;
double error;
//LTimer t; t.start();
distances.clear();
indexes.clear();
#ifndef USE_KD
Mat srcm = asMatRows(src), dstm = asMatRows(dst);
vector<float> dists;
error = (double) knn(dstm, srcm, indexes, dists);
for(int i = 0; i < dists.size(); i++) distances.push_back((double)dists[i]);
double divisor = src.size() > 0 ? (double)(src.size()) : 1.0;
error /= divisor;
#else
LKDNode n; n.init(dst);
n.split(dst, 0, 15);
//cout << n.averageLeafSize() << endl;
error = 0.0;
for(int i = 0; i < src.size(); i++)
{
R3 p = src[i];
R3 w;
int index = 0;
n.NN(dst,p, index, w);
//if(i%1000 == 0 )cout << i << " / " << src.size() << endl;
indexes.push_back(index);
float D = p.dist(dst[index]);
distances.push_back(D);
error += (D*D) / (float) src.size();
}
#endif
//t.stop();
//cout << "took " << t.getSeconds() << endl;
return error;
}
void main()
{
int movie,rating,ex,usr,user,cnt,ind;
for(i=0;i<100000;i++)
{
scanf("%d%d%d%d",&usr,&movie,&rating,&ex);
r[usr-1][movie-1]=rating;
}
root=knn(root);
//display();
node ptr;
ptr=root;
user=2;
printf("the recommended items for %d are:\n",user);
while(1)
{
if(ptr->head==user)
break;
ptr=ptr->next;
}
for(j=0;j<1682;j++)
{
cnt=0;
for(k=0;k<10;k++)
{
ind=ptr->array[k];
if((r[ind][j]!=0)&&(r[user][j]==0))
{
cnt++;
}
}
if(cnt>6)
printf("%d\t",j);
}
}
/* NEW VERSION
*/
void linsched_create_normal_task_binary(struct elf_binary* binary, int niceval)
{
struct sched_param params;
unsigned long long time_slice;
int query[3];
query[0] = binary->size;
query[1] = binary->bss;
query[2] = binary->input_size;
/* If we cannot support any more tasks, return. */
if (curr_task_id >= LINSCHED_MAX_TASKS)
return;
time_start = sched_clock();
sim_param = (query[0]+query[1]+query[2])/(30*2);
time_slice = knn(&query[0], 3);
time_slice *= 1000000;
//time_slice = 500000000;
/* Create "normal" task and set its nice value. */
__linsched_tasks[curr_task_id] = __linsched_create_task_binary(binary->callback, time_slice);
params.sched_priority = 0;
sys_sched_setscheduler(__linsched_tasks[curr_task_id], SCHED_NORMAL,
¶ms);
set_user_nice(__linsched_tasks[curr_task_id], niceval);
/* Print message. */
printf("Created normal task 0x%x with nice value %d.\n",
(unsigned int)__linsched_tasks[curr_task_id], niceval);
/* Increment task id. */
curr_task_id++;
}
int main(int argc, char** argv)
{
CoverTree<float, std::vector<float>, float (*const)(const std::vector<float>&, const std::vector<float>&)> tree(&euclidian);
PointContainer data = generate(VECTORLENGTH);
boost::posix_time::ptime time = boost::posix_time::microsec_clock::local_time();
for(PointContainer::const_iterator it = data.begin(); it != data.end(); ++it)
{
tree.insert(*it);
}
std::cout << "Build time " << (boost::posix_time::microsec_clock::local_time() - time) << std::endl;
std::ofstream stream("dump.txt");
tree.dump(stream);
Point zero(2, 0.f);
time = boost::posix_time::microsec_clock::local_time();
PointContainer result = knn(data, zero, KNNSIZE);
std::cout << "Out time (linear) " << (boost::posix_time::microsec_clock::local_time() - time) << std::endl;
time = boost::posix_time::microsec_clock::local_time();
PointContainer result2 = tree.knn(zero, KNNSIZE);
std::cout << "Out time (cover_tree) " << (boost::posix_time::microsec_clock::local_time() - time) << std::endl;
bool boolean = true;
for(int i = 0; i < std::min(result.size(), result2.size()); ++i)
{
for(int j = 0; j < POINTSIZE; ++j)
{
boolean = boolean && (result[i][j] == result2[i][j]);
}
std::cout << i << std::endl << result[i] << result2[i];
}
std::cout << "Result " << (boolean && (result.size() == result2.size())) << std::endl;
return EXIT_SUCCESS;
}
/* Compute the nearest neighbors */
void SMOTE::computeNeighbors(cv::Mat minority, int nearestNeighbors,
cv::Mat *neighbors) {
cv::Size s = minority.size();
int i, k = nearestNeighbors, max, index = 0;
cv::Mat classes(s.height, 1, CV_32FC1);
for (i = 0; i < s.height; i++) {
classes.at<float>(i, 0) = index;
index++;
}
cv::KNearest knn(minority, classes);
max = (minority.rows > knn.get_max_k() ? knn.get_max_k() : minority.rows);
knn.find_nearest(minority, (k > max ? max : k), 0, 0, neighbors, 0);
// for (i = 0; i < (*neighbors).size().height; i++){
// std::cout << std::endl << nearestNeighbors << " vizinhos de i: " << i << std::endl;
// for (int x = 0; x < (*neighbors).size().width; x++){
// std::cout << (*neighbors).at<float>(i, x) << " ";
// }
// }
classes.release();
}
void Apply_knn:: getAll_knn(int sizeOfMat,std::vector<float> label_bucket[],std::vector<float> values_of_bucket[])
{
/**
* Begin Apply knn to the Mat
*
**/
std::vector<float> container[sizeOfMat];
int k=15;
cv::Mat result=Mat(sizeOfMat,k,CV_32F);
/**
* Load data for knn
**/
CvKNearest knn(&trainningMat,&labelMat);
/**
* Loading complet
**/
knn.find_nearest(&testingMat,k,&result);
cout<<"get max_k\t"<<knn.get_max_k()<<"\n";
cout<<"get total trainn samples\t"<<knn.get_sample_count()<<"\n";
cout<<result;
cout<<"\n";
/**
* End Apply knn to the Mat
*
**/
/**
* Put values in the bucket appropriately begins
*
**/
cv::Mat tempMat=cv::Mat(&testingMat);
for(int _i=0;_i<result.rows;_i++)
{
/**
* _i points to the first row of result which corresponds to the first row of testing sample.
* So points at testing sample of _i must be put to the same index of index_to_bucket
**/
for(int _j=0;_j<result.cols;_j++)
{
int key_of_map=result.at<float>(_i);/**here will be change jth index needs to be added. So check cols of result**/
int index_to_bucket=vect_indexMap[key_of_map];
cout<<index_to_bucket<<"\t"<<key_of_map<<"\n";
/** index of testing sample serially from begin to end*/
int index_to_testingSample=testingIndexMat.at<int>(_i);
(label_bucket[index_to_bucket]).push_back(index_to_testingSample);
float _x=tempMat.at<float>(_i,0);
float _y=tempMat.at<float>(_i,1);
(values_of_bucket[index_to_bucket]).push_back(_x);
(values_of_bucket[index_to_bucket]).push_back(_y);
}/**check this again**/
}
/**
* Put values in the bucket appropriately ends
*
**/
/**
* Printing the values in the buckets begin
*
**/
for(int _j=0;_j<result.rows-1;_j++)
{
//std::vector<float>::iterator it_to_label_bucket;
cout<<"Contains of bucket:\t"<<(label_bucket[_j]).at(0)<<"\n";
for(int _k=0;_k<(label_bucket[_j]).size();_k++)
{
cout<<(label_bucket[_j]).at(_k)<<"\t:";
cout<<"_x\t:"<<(values_of_bucket[_j]).at(_k*2)<<"\t";
cout<<"_y\t:"<<(values_of_bucket[_j]).at(_k*2+1);
cout<<"\n";
}
cout<<"\n\n";
}
/**
* Printing the values in the buckets end
*
**/
WriteGroupFile wgrF;
wgrF.setFileName("gr_1nnD.txt");
wgrF.openFile();
wgrF.imageNumber=((label_bucket[0]).at(0))/500+1;
if(wgrF.imageNumber==1)
{
/**
* Saving all contents of lable_bucket in
**/
cout<<"here\n";
int total_groups=0;
if(result.rows%2==0)
total_groups=result.rows;
else
total_groups=result.rows-1;
cout<<"Total groups:\t"<<total_groups<<"\n";
getFirstImageVector(label_bucket,total_groups);
}
wgrF.setTotalBuckets(sizeOfMat);
wgrF.writeGroups(label_bucket,values_of_bucket);
wgrF.closeFile();
// return listOfMat;
}
static std::pair<int,int> leave_one_out(KnnObject* o, int stop_threshold,
int* selection_vector = 0,
double* weight_vector = 0,
std::vector<long>* indexes = 0) {
int* selections = selection_vector;
if (selections == 0) {
selections = o->selection_vector;
}
double* weights = weight_vector;
if (weights == 0) {
weights = o->weight_vector;
}
assert(o->feature_vectors != 0);
kNearestNeighbors<char*, ltstr, eqstr> knn(o->num_k);
int total_correct = 0;
int total_queries = 0;
if (indexes == 0) {
for (size_t i = 0; i < o->feature_vectors->size(); ++i/*, cur += o->num_features*/) {
// We don't want to do the calculation if there is no
// hope that kNN will return the correct answer (because
// there aren't enough examples in the database).
if (o->id_name_histogram[i] < int((o->num_k + 0.5) / 2)) {
continue;
}
double* current_known;
double* unknown = (*o->feature_vectors)[i];
for (size_t j = 0; j < o->feature_vectors->size(); ++j) {
current_known = (*o->feature_vectors)[j];
if (i == j)
continue;
double distance;
compute_distance(o->distance_type, current_known, o->num_features,
unknown, &distance, selections, weights);
knn.add(o->id_names[j], distance);
}
knn.majority();
if (strcmp(knn.answer[0].first, o->id_names[i]) == 0) {
total_correct++;
}
knn.reset();
total_queries++;
if (total_queries - total_correct > stop_threshold)
return std::make_pair(total_correct, total_queries);
}
} else {
for (size_t i = 0; i < o->feature_vectors->size(); ++i) {
if (o->id_name_histogram[i] < int((o->num_k + 0.5) / 2))
continue;
double* current_known;
double* unknown = (*o->feature_vectors)[i];
for (size_t j = 0; j < o->feature_vectors->size(); ++j) {
current_known = (*o->feature_vectors)[j];
if (i == j)
continue;
double distance;
if (o->distance_type == CITY_BLOCK) {
distance = city_block_distance_skip(current_known, unknown, selections, weights,
indexes->begin(), indexes->end());
} else if (o->distance_type == FAST_EUCLIDEAN) {
distance = fast_euclidean_distance_skip(current_known, unknown, selections, weights,
indexes->begin(), indexes->end());
} else {
distance = euclidean_distance_skip(current_known, unknown, selections, weights,
indexes->begin(), indexes->end());
}
knn.add(o->id_names[j], distance);
}
knn.majority();
if (strcmp(knn.answer[0].first, o->id_names[i]) == 0) {
total_correct++;
}
knn.reset();
total_queries++;
if (total_queries - total_correct > stop_threshold)
return std::make_pair(total_correct, total_queries);
}
}
return std::make_pair(total_correct, total_queries);
}
void mexFunction(int nout, mxArray *pout[], int nin, const mxArray *pin[]) {
FILE *f = fopen("log.txt", "wt");
fclose(f);
if (nin < 2) { mexErrMsgTxt("nnmex called with < 2 input arguments"); }
const mxArray *A = pin[0], *B = pin[1];
const mxArray *ANN_PREV = NULL, *ANN_WINDOW = NULL, *AWINSIZE = NULL;
int aw = -1, ah = -1, bw = -1, bh = -1;
BITMAP *a = NULL, *b = NULL, *ann_prev = NULL, *ann_window = NULL, *awinsize = NULL;
VECBITMAP<unsigned char> *ab = NULL, *bb = NULL;
VECBITMAP<float> *af = NULL, *bf = NULL;
if (mxGetNumberOfDimensions(A) != 3 || mxGetNumberOfDimensions(B) != 3) { mexErrMsgTxt("dims != 3"); }
if (mxGetDimensions(A)[2] != mxGetDimensions(B)[2]) { mexErrMsgTxt("3rd dimension not same size"); }
int mode = MODE_IMAGE;
if (mxGetDimensions(A)[2] != 3) { // a discriptor rather than rgb
if (mxIsUint8(A) && mxIsUint8(B)) { mode = MODE_VECB; }
else if (is_float(A) && is_float(B)) { mode = MODE_VECF; }
else { mexErrMsgTxt("input not uint8, single, or double"); }
}
Params *p = new Params();
RecomposeParams *rp = new RecomposeParams();
BITMAP *borig = NULL;
if (mode == MODE_IMAGE) {
a = convert_bitmap(A);
b = convert_bitmap(B);
borig = b;
aw = a->w; ah = a->h;
bw = b->w; bh = b->h;
}
else if (mode == MODE_VECB) {
ab = convert_vecbitmap<unsigned char>(A);
bb = convert_vecbitmap<unsigned char>(B);
if (ab->n != bb->n) { mexErrMsgTxt("3rd dimension differs"); }
aw = ab->w; ah = ab->h;
bw = bb->w; bh = bb->h;
p->vec_len = ab->n;
}
else if (mode == MODE_VECF) {
af = convert_vecbitmap<float>(A);
bf = convert_vecbitmap<float>(B);
if (af->n != bf->n) { mexErrMsgTxt("3rd dimension differs"); }
aw = af->w; ah = af->h;
bw = bf->w; bh = bf->h;
p->vec_len = af->n;
}
double *win_size = NULL;
BITMAP *amask = NULL, *bmask = NULL;
double scalemin = 0.5, scalemax = 2.0; // The product of these must be one.
/* parse parameters */
int i = 2;
int sim_mode = 0;
int knn_chosen = -1;
p->algo = ALGO_CPU;
int enrich_mode = 0;
if (nin > i && !mxIsEmpty(pin[i])) {
if (mxStringEquals(pin[i], "cpu")) { p->algo = ALGO_CPU; }
else if (mxStringEquals(pin[i], "gpucpu")) { p->algo = ALGO_GPUCPU; }
else if (mxStringEquals(pin[i], "cputiled")) { p->algo = ALGO_CPUTILED; }
else if (mxStringEquals(pin[i], "rotscale")) { sim_mode = 1; }
else if (mxStringEquals(pin[i], "enrich")) { p->algo = ALGO_CPUTILED; enrich_mode = 1; }
else { mexErrMsgTxt("Unknown algorithm"); }
} i++;
if (nin > i && !mxIsEmpty(pin[i])) { p->patch_w = int(mxGetScalar(pin[i])); } i++;
if (nin > i && !mxIsEmpty(pin[i])) { p->nn_iters = int(mxGetScalar(pin[i])); } i++;
if (nin > i && !mxIsEmpty(pin[i])) { p->rs_max = int(mxGetScalar(pin[i])); } i++;
if (nin > i && !mxIsEmpty(pin[i])) { p->rs_min = int(mxGetScalar(pin[i])); } i++;
if (nin > i && !mxIsEmpty(pin[i])) { p->rs_ratio = mxGetScalar(pin[i]); } i++;
if (nin > i && !mxIsEmpty(pin[i])) { p->rs_iters = mxGetScalar(pin[i]); } i++;
if (nin > i && !mxIsEmpty(pin[i])) { p->cores = int(mxGetScalar(pin[i])); } i++;
if (nin > i && !mxIsEmpty(pin[i])) { bmask = convert_bitmap(pin[i]); } i++; // XC+
if (nin > i && !mxIsEmpty(pin[i])) {
if (!mxIsDouble(pin[i])) { mexErrMsgTxt("\nwin_size should be of type double."); }
win_size = (double*)mxGetData(pin[i]);
if (mxGetNumberOfElements(pin[i])==1) { p->window_h = p->window_w = int(win_size[0]); }
else if (mxGetNumberOfElements(pin[i])==2) { p->window_h = int(win_size[0]); p->window_w = int(win_size[1]); }
else { mexErrMsgTxt("\nwin_size should be a scalar for square window or [h w] for a rectangular one."); }
} i++;
/* continue parsing parameters */
// [ann_prev=NULL], [ann_window=NULL], [awinsize=NULL],
if (nin > i && !mxIsEmpty(pin[i])) {
ANN_PREV = pin[i];
int clip_count = 0;
ann_prev = convert_field(p, ANN_PREV, bw, bh, clip_count); // Bug fixed by Connelly
} i++;
if (nin > i && !mxIsEmpty(pin[i])) {
ANN_WINDOW = pin[i];
int clip_count = 0;
ann_window = convert_field(p, ANN_WINDOW, bw, bh, clip_count);
} i++;
if (nin > i && !mxIsEmpty(pin[i])) {
AWINSIZE = pin[i];
awinsize = convert_winsize_field(p, AWINSIZE, aw, ah);
if (p->window_w==INT_MAX||p->window_h==INT_MAX) { p->window_w = -1; p->window_h = -1; }
} i++;
if (nin > i && !mxIsEmpty(pin[i])) {
knn_chosen = int(mxGetScalar(pin[i]));
if (knn_chosen == 1) { knn_chosen = -1; }
if (knn_chosen <= 0) { mexErrMsgTxt("\nknn is less than zero"); }
} i++;
if (nin > i && !mxIsEmpty(pin[i])) {
scalemax = mxGetScalar(pin[i]);
if (scalemax <= 0) { mexErrMsgTxt("\nscalerange is less than zero"); }
scalemin = 1.0/scalemax;
if (scalemax < scalemin) {
double temp = scalemax;
scalemax = scalemin;
scalemin = temp;
}
} i++;
if (ann_window&&!awinsize&&!win_size) {
mexErrMsgTxt("\nUsing ann_window - either awinsize or win_size should be defined.\n");
}
if (enrich_mode) {
int nn_iters = p->nn_iters;
p->enrich_iters = nn_iters/2;
p->nn_iters = 2;
if (A != B) { mexErrMsgTxt("\nOur implementation of enrichment requires that image A = image B.\n"); }
if (mode == MODE_IMAGE) {
b = a;
} else {
mexErrMsgTxt("\nEnrichment only implemented for 3 channel uint8 inputs.\n");
}
}
init_params(p);
if (sim_mode) {
init_xform_tables(scalemin, scalemax, 1);
}
RegionMasks *amaskm = amask ? new RegionMasks(p, amask): NULL;
BITMAP *ann = NULL; // NN field
BITMAP *annd_final = NULL; // NN patch distance field
BITMAP *ann_sim_final = NULL;
VBMP *vann_sim = NULL;
VBMP *vann = NULL;
VBMP *vannd = NULL;
if (mode == MODE_IMAGE) {
// input as RGB image
if (!a || !b) { mexErrMsgTxt("internal error: no a or b image"); }
if (knn_chosen > 1) {
p->knn = knn_chosen;
if (sim_mode) { mexErrMsgTxt("rotating+scaling patches not implemented with knn (actually it is implemented it is not exposed by the wrapper)"); }
PRINCIPAL_ANGLE *pa = NULL;
vann_sim = NULL;
vann = knn_init_nn(p, a, b, vann_sim, pa);
vannd = knn_init_dist(p, a, b, vann, vann_sim);
knn(p, a, b, vann, vann_sim, vannd, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, pa);
// sort_knn(p, vann, vann_sim, vannd);
} else if (sim_mode) {
BITMAP *ann_sim = NULL;
ann = sim_init_nn(p, a, b, ann_sim);
BITMAP *annd = sim_init_dist(p, a, b, ann, ann_sim);
sim_nn(p, a, b, ann, ann_sim, annd);
if (ann_prev) { mexErrMsgTxt("when searching over rotations+scales, previous guess is not supported"); }
annd_final = annd;
ann_sim_final = ann_sim;
} else {
ann = init_nn(p, a, b, bmask, NULL, amaskm, 1, ann_window, awinsize);
BITMAP *annd = init_dist(p, a, b, ann, bmask, NULL, amaskm);
nn(p, a, b, ann, annd, amaskm, bmask, 0, 0, rp, 0, 0, 0, NULL, p->cores, ann_window, awinsize);
if (ann_prev) minnn(p, a, b, ann, annd, ann_prev, bmask, 0, 0, rp, NULL, amaskm, p->cores);
annd_final = annd;
}
}
/*
else if (mode == MODE_VECB) {
// mexPrintf("mode vecb %dx%dx%d, %dx%dx%d\n", ab->w, ab->h, ab->n, bb->w, bb->h, bb->n);
// mexPrintf(" %d %d %d %d\n", ab->get(0,0)[0], ab->get(1,0)[0], ab->get(0,1)[0], ab->get(0,0)[1]);
if (!ab || !bb) { mexErrMsgTxt("internal error: no a or b image"); }
ann = vec_init_nn<unsigned char>(p, ab, bb, bmask, NULL, amaskm);
VECBITMAP<int> *annd = vec_init_dist<unsigned char, int>(p, ab, bb, ann, bmask, NULL, amaskm);
// mexPrintf(" %d %d %d %p %p\n", annd->get(0,0)[0], annd->get(1,0)[0], annd->get(0,1)[0], amaskm, bmask);
vec_nn<unsigned char, int>(p, ab, bb, ann, annd, amaskm, bmask, 0, 0, rp, 0, 0, 0, NULL, p->cores);
if (ann_prev) vec_minnn<unsigned char, int>(p, ab, bb, ann, annd, ann_prev, bmask, 0, 0, rp, NULL, amaskm, p->cores);
annd_final = vecbitmap_to_bitmap(annd);
delete annd;
}
else if (mode == MODE_VECF) {
// mexPrintf("mode vecf %dx%dx%d, %dx%dx%d\n", af->w, af->h, af->n, bf->w, bf->h, bf->n);
// mexPrintf(" %f %f %f %f\n", af->get(0,0)[0], af->get(1,0)[0], af->get(0,1)[0], af->get(0,0)[1]);
if (!af || !bf) { mexErrMsgTxt("internal error: no a or b image"); }
ann = vec_init_nn<float>(p, af, bf, bmask, NULL, amaskm);
VECBITMAP<float> *annd = vec_init_dist<float, float>(p, af, bf, ann, bmask, NULL, amaskm);
vec_nn<float, float>(p, af, bf, ann, annd, amaskm, bmask, 0, 0, rp, 0, 0, 0, NULL, p->cores);
if (ann_prev) vec_minnn<float, float>(p, af, bf, ann, annd, ann_prev, bmask, 0, 0, rp, NULL, amaskm, p->cores);
annd_final = create_bitmap(annd->w, annd->h);
clear(annd_final);
delete annd;
}
*/
else if(mode == MODE_VECB) {
if (sim_mode) { mexErrMsgTxt("internal error: rotation+scales not implemented with descriptor mode"); }
if (knn_chosen > 1) { mexErrMsgTxt("internal error: kNN not implemented with descriptor mode"); }
// mexPrintf("mode vecb_xc %dx%dx%d, %dx%dx%d\n", ab->w, ab->h, ab->n, bb->w, bb->h, bb->n);
// input as uint8 discriptors per pixel
if (!ab || !bb) { mexErrMsgTxt("internal error: no a or b image"); }
ann = XCvec_init_nn<unsigned char>(p, ab, bb, bmask, NULL, amaskm);
VECBITMAP<int> *annd = XCvec_init_dist<unsigned char, int>(p, ab, bb, ann, bmask, NULL, amaskm);
XCvec_nn<unsigned char, int>(p, ab, bb, ann, annd, amaskm, bmask, 0, 0, rp, 0, 0, 0, NULL, p->cores);
if (ann_prev) XCvec_minnn<unsigned char, int>(p, ab, bb, ann, annd, ann_prev, bmask, 0, 0, rp, NULL, amaskm, p->cores);
annd_final = vecbitmap_to_bitmap(annd);
delete annd;
} else if(mode == MODE_VECF) {
if (sim_mode) { mexErrMsgTxt("internal error: rotation+scales not implemented with descriptor mode"); }
if (knn_chosen > 1) { mexErrMsgTxt("internal error: kNN not implemented with descriptor mode"); }
// input as float/double discriptors per pixel
if (!af || !bf) { mexErrMsgTxt("internal error: no a or b image"); }
ann = XCvec_init_nn<float>(p, af, bf, bmask, NULL, amaskm);
VECBITMAP<float> *annd = XCvec_init_dist<float, float>(p, af, bf, ann, bmask, NULL, amaskm);
XCvec_nn<float, float>(p, af, bf, ann, annd, amaskm, bmask, 0, 0, rp, 0, 0, 0, NULL, p->cores);
if (ann_prev) XCvec_minnn<float, float>(p, af, bf, ann, annd, ann_prev, bmask, 0, 0, rp, NULL, amaskm, p->cores);
annd_final = create_bitmap(annd->w, annd->h);
clear(annd_final);
delete annd;
}
destroy_region_masks(amaskm);
// output ann: x | y | patch_distance
if(nout >= 1) {
mxArray *ans = NULL;
if (knn_chosen > 1) {
if (sim_mode) { mexErrMsgTxt("rotating+scaling patches return value not implemented with knn"); }
mwSize dims[4] = { ah, aw, 3, knn_chosen };
ans = mxCreateNumericArray(4, dims, mxINT32_CLASS, mxREAL);
int *data = (int *) mxGetData(ans);
for (int kval = 0; kval < knn_chosen; kval++) {
int *xchan = &data[aw*ah*3*kval+0];
int *ychan = &data[aw*ah*3*kval+aw*ah];
int *dchan = &data[aw*ah*3*kval+2*aw*ah];
for (int y = 0; y < ah; y++) {
// int *ann_row = (int *) ann->line[y];
// int *annd_row = (int *) annd_final->line[y];
for (int x = 0; x < aw; x++) {
// int pp = ann_row[x];
int pp = vann->get(x, y)[kval];
int pos = y + x * ah;
xchan[pos] = INT_TO_X(pp);
ychan[pos] = INT_TO_Y(pp);
dchan[pos] = vannd->get(x, y)[kval];
}
}
}
} else if (ann_sim_final) {
mwSize dims[3] = { ah, aw, 5 };
ans = mxCreateNumericArray(3, dims, mxSINGLE_CLASS, mxREAL);
float *data = (float *) mxGetData(ans);
float *xchan = &data[0];
float *ychan = &data[aw*ah];
float *dchan = &data[2*aw*ah];
float *tchan = &data[3*aw*ah];
float *schan = &data[4*aw*ah];
double angle_scale = 2.0*M_PI/NUM_ANGLES;
for (int y = 0; y < ah; y++) {
int *ann_row = (int *) ann->line[y];
int *annd_row = (int *) annd_final->line[y];
int *ann_sim_row = ann_sim_final ? (int *) ann_sim_final->line[y]: NULL;
for (int x = 0; x < aw; x++) {
int pp = ann_row[x];
int pos = y + x * ah;
xchan[pos] = INT_TO_X(pp);
ychan[pos] = INT_TO_Y(pp);
dchan[pos] = annd_row[x];
if (ann_sim_final) {
int v = ann_sim_row[x];
int tval = INT_TO_Y(v)&(NUM_ANGLES-1);
int sval = INT_TO_X(v);
tchan[pos] = tval*angle_scale;
schan[pos] = xform_scale_table[sval]*(1.0/65536.0);
}
}
}
} else {
mwSize dims[3] = { ah, aw, 3 };
ans = mxCreateNumericArray(3, dims, mxINT32_CLASS, mxREAL);
int *data = (int *) mxGetData(ans);
int *xchan = &data[0];
int *ychan = &data[aw*ah];
int *dchan = &data[2*aw*ah];
for (int y = 0; y < ah; y++) {
int *ann_row = (int *) ann->line[y];
int *annd_row = (int *) annd_final->line[y];
for (int x = 0; x < aw; x++) {
int pp = ann_row[x];
int pos = y + x * ah;
xchan[pos] = INT_TO_X(pp);
ychan[pos] = INT_TO_Y(pp);
dchan[pos] = annd_row[x];
}
}
}
pout[0] = ans;
}
// clean up
delete vann;
delete vann_sim;
delete vannd;
delete p;
delete rp;
destroy_bitmap(a);
destroy_bitmap(borig);
delete ab;
delete bb;
delete af;
delete bf;
destroy_bitmap(ann);
destroy_bitmap(annd_final);
destroy_bitmap(ann_sim_final);
if (ann_prev) destroy_bitmap(ann_prev);
if (ann_window) destroy_bitmap(ann_window);
if (awinsize) destroy_bitmap(awinsize);
}
int main (){
char file[50];//name of file
int nclus;//number of clusters
//float *Cone, *Ctwo, *temp, *temp2;
/*--------------------------------------------------------------------------
-------------get input parameters
--------------------------------------------------------------------------*/
//get name of file to open
printf("what is the name of your file?");
if (scanf("%s", &file[0]) != 1){ //read in value and make sure it is valid name
printf("error: filemane invalid!\n");
exit(EXIT_FAILURE);
}
printf("filename is %s\n", file);
//get number of clusters
printf("how many clusters do you want to find?");
if (scanf("%d", &nclus) != 1){ //read in value and make sure it is valid number
printf("error: number invalid!\n");
exit(EXIT_FAILURE);
}
printf("number of k clusters is %d\n", nclus);
/*--------------------------------------------------------------------------
-------------open and read input file
--------------------------------------------------------------------------*/
//open input file
FILE *fp;
if ((fp = fopen(file,"r")) == NULL){
printf("Error:cannot read the file %s\n", file);
}else{
printf("success opening file %s\n", file);
}
//read number of items and attributes
int items, attributes;
fscanf(fp,"%d",&items);
fscanf(fp,"%d",&attributes);
printf ("items %d attributes %d\n", items, attributes);
getchar();
//Create nodes
Node *head = (Node *) malloc (sizeof(Node));
Node *node = (Node *) malloc (sizeof(Node));
float *temp = (float *)malloc(items*attributes*sizeof(float));
//read file
printf("reading file\n");
int i, j; //counters
for (i=0; i < items; i++){//for each item
printf("reading item %d\n",i);
for (j=0; j < attributes; j++){
fscanf(fp,"%f",&temp[(attributes)*i+j]);
//printf("read attribute %d, %f\n",j,temp[(attributes)*i+j]);
}
node = addData(nclus+1,&temp[(attributes)*i]);
append(head,node);
}
/*--------------------------------------------------------------------------
-------------get initial means
--------------------------------------------------------------------------*/
float *means = (float *)malloc(nclus*attributes*sizeof(float));
//float *means_b = (float *)malloc(nclus*attributes*sizeof(float));
int index;
node = head;
//printf("node...%f %f\n", node->att[0], node->att[1]);
for (i=0;i<nclus;i++){
node = node->next;
//printf("node...%f %f\n", node->att[0], node->att[1]);
printf("\n\ninitial center for cluster %d:\n",i);
for (j=0;j<attributes;j++){
index = (attributes)*(i)+j;
means[index] = node->att[j];
printf("attribute %d: %f\n", j, node->att[j]);
}
}
printf("press enter\n");
getchar();
/*--------------------------------------------------------------------------
-------------run k-means algorithm
--------------------------------------------------------------------------*/
//declare variables for k-means algo
float *att_sum = (float *)malloc(nclus*attributes*sizeof(float)); //sum of each column for each cluster
float *sqerr_sum = (float *)malloc(nclus*sizeof(float)); //keep track of rmse
float *rmse = (float *)malloc(nclus*sizeof(float)); //keep track of rmse
float *rmsediff = (float *)malloc(nclus*sizeof(float)); //differences in rmse
int *cnt = (int *)malloc(nclus*sizeof(int)); //number of elements in each cluster
float *last_rmse = (float *)malloc(nclus*sizeof(float));
int test = 0;
int count = 0;
do{//repeat until test = 1 (RMSE converges)
test = kmeans(head,&means[0],nclus,&last_rmse[0], attributes,
&att_sum[0], &sqerr_sum[0], &rmse[0], &rmsediff[0], &cnt[0]);
printf("press enter for next iteration\n");
getchar();
//printf("test %d\n", test);
count ++;
if (count > 100)
test = 1;
}while(test == 0);
/*--------------------------------------------------------------------------
-------------print results in output file
--------------------------------------------------------------------------*/
FILE *out;
if ((out = fopen("output_gbm.csv","w")) == NULL){
printf("Error:cannot read the file output.txt\n");
}else{
printf("success opening output file output.txt for output\n");
}
node = head;
while(node->next != NULL){
node = node->next;
for (j=0;j<attributes;j++) {
fprintf(out, "%f,", node->att[j]);
}
fprintf(out,"%d\n",node->kclus);
}
fclose(out);
fclose (fp) ;
/*--------------------------------------------------------------------------
-------------run k-nearest neighbor algorithm
--------------------------------------------------------------------------*/
int response;
printf("if you want to run k-nearest neighbor type the number 1 and press enter, otherwise press any other key: ");
if (scanf("%d", &response) != 1){ //read in value and make sure it is valid number
printf("error: number invalid!\n");
exit(EXIT_FAILURE);
}
printf("response is %d\n", response);
//run k-nn algorithm
if (response == 1){
knn(head, nclus); //knn call function
}
free(last_rmse);
free(means);
free(att_sum);
free(sqerr_sum);
free(rmse);
free(rmsediff);
free(cnt);
return 1;
}
bool ccGriddedTools::ComputeNormals(ccPointCloud* cloud,
const std::vector<int>& indexGrid,
int width,
int height,
bool* canceledByUser/*=0*/,
int kernelWidth/*=3*/ )
{
//init
bool result = true;
if (canceledByUser)
*canceledByUser = false;
//try to compute normals
if (cloud->reserveTheNormsTable())
{
//progress dialog
ccProgressDialog pdlg(true);
CCLib::NormalizedProgress nprogress(&pdlg,cloud->size());
pdlg.setMethodTitle("Compute normals");
pdlg.setInfo(qPrintable(QString("Number of points: %1").arg(cloud->size())));
pdlg.start();
const int* _indexGrid = &(indexGrid[0]);
CCLib::ReferenceCloud knn(cloud);
//neighborhood 'half-width' (total width = 1 + 2*kernelWidth)
//max number of neighbours: (1+2*nw)^2
knn.reserve((1+2*kernelWidth)*(1+2*kernelWidth));
//for each grid cell
for (int j=0; j<height; ++j)
{
for (int i=0; i<width; ++i, ++_indexGrid)
{
if (*_indexGrid >= 0)
{
unsigned pointIndex = static_cast<unsigned>(*_indexGrid);
//add the point itself
knn.clear(false);
knn.addPointIndex(pointIndex); //warning: indexes are shifted (0 = no point)
const CCVector3* P = cloud->getPoint(pointIndex);
//look for neighbors
for (int v=std::max(0,j-kernelWidth); v<std::min<int>(height,j+kernelWidth); ++v)
{
if (v != j)
{
for (int u=std::max(0,i-kernelWidth); u<std::min<int>(width,i+kernelWidth); ++u)
{
if (u != i)
{
int indexN = indexGrid[v*width + u];
if (indexN >= 0)
{
//we don't consider points with a too much different depth than the central point
const CCVector3* Pn = cloud->getPoint(static_cast<unsigned>(indexN));
if (fabs(Pn->z - P->z) <= std::max(fabs(Pn->x - P->x),fabs(Pn->y - P->y)))
knn.addPointIndex(static_cast<unsigned>(indexN)); //warning: indexes are shifted (0 = no point)
}
}
}
}
}
CCVector3 N(0,0,1);
if (knn.size() >= 3)
{
CCLib::Neighbourhood Z(&knn);
//compute normal with quadratic func. (if we have enough points)
if (false/*knn.size() >= 6*/)
{
uchar hfDims[3];
const PointCoordinateType* h = Z.getHeightFunction(hfDims);
if (h)
{
const CCVector3* gv = Z.getGravityCenter();
assert(gv);
const uchar& iX = hfDims[0];
const uchar& iY = hfDims[1];
const uchar& iZ = hfDims[2];
PointCoordinateType lX = P->u[iX] - gv->u[iX];
PointCoordinateType lY = P->u[iY] - gv->u[iY];
N.u[iX] = h[1] + (2 * h[3] * lX) + (h[4] * lY);
N.u[iY] = h[2] + (2 * h[5] * lY) + (h[4] * lX);
N.u[iZ] = -PC_ONE;
N.normalize();
}
}
else
#define USE_LSQ_PLANE
#ifdef USE_LSQ_PLANE
{
//compute normal with best fit plane
const CCVector3* _N = Z.getLSQPlaneNormal();
if (_N)
N = *_N;
}
#else
{
//compute normals with 2D1/2 triangulation
CCLib::GenericIndexedMesh* theMesh = Z.triangulateOnPlane();
if (theMesh)
{
unsigned faceCount = theMesh->size();
N.z = 0;
//for all triangles
theMesh->placeIteratorAtBegining();
for (unsigned j=0; j<faceCount; ++j)
{
const CCLib::TriangleSummitsIndexes* tsi = theMesh->getNextTriangleIndexes();
//we look if the central point is one of the triangle's vertices
if (tsi->i1 == 0 || tsi->i2 == 0|| tsi->i3 == 0)
{
const CCVector3 *A = knn.getPoint(tsi->i1);
const CCVector3 *B = knn.getPoint(tsi->i2);
const CCVector3 *C = knn.getPoint(tsi->i3);
CCVector3 no = (*B - *A).cross(*C - *A);
//no.normalize();
N += no;
}
}
delete theMesh;
theMesh = 0;
//normalize the 'mean' vector
N.normalize();
}
}
#endif
}
//check normal vector sign
CCVector3 viewVector = *P /*- cloudTrans.getTranslationAsVec3D()*/; //clouds are still in their local coordinate system!
if (viewVector.dot(N) > 0)
N *= -PC_ONE;
cloud->addNorm(N);
//progress
if (!nprogress.oneStep())
{
result = false;
if (canceledByUser)
*canceledByUser = true;
ccLog::Warning("[ComputeNormals] Process canceled by user!");
//early stop
j = height;
break;
}
}
}
}
if (!result)
{
cloud->unallocateNorms();
}
}
else
{
ccLog::Warning("[ComputeNormals] Not enough memory!");
}
return result;
}
float knn(Mat& m_destinations, Mat& m_object, vector<int>& ptpairs)
{
vector<float> dists;
return knn(m_destinations, m_object, ptpairs, dists);
}
int main (int argc, const char * argv[])
{
// choose which zero of autocorrelation should be set as delay parameter
int delay_choice =2;
int embed_dimension =3;
int num_neighbours = 15;
long num_of_points_retained=500;
int percentile =70;
long output_len=2*(((float)percentile)/100)*num_of_points_retained;
printf("%ld output_len =",output_len);
float *output = malloc(output_len* sizeof(float));
printf("Density based subsampling running \n");
if (argc != 2) {
fprintf(stderr, "Expecting wav file as argument\n");
return 1;
}
// Open sound file that comes in as a command line argument
SF_INFO sndInfo;
SNDFILE *sndFile = sf_open(argv[1], SFM_READ, &sndInfo);
if (sndFile == NULL) {
fprintf(stderr, "Error reading source file '%s': %s\n", argv[1], sf_strerror(sndFile));
return 1;
}
// Check format - 16bit PCM
if (sndInfo.format != (SF_FORMAT_WAV | SF_FORMAT_PCM_16)) {
fprintf(stderr, "Input should be 16bit Wav\n");
sf_close(sndFile);
return 1;
}
// Now we know the length of the wav file, we can create a buffer for it, in this case, we are creating a double array.
// Allocate memory
float *buffer = malloc(sndInfo.frames * sizeof(float)* sndInfo.channels);
if (buffer == NULL)
{
fprintf(stderr, "Could not allocate memory for data\n");
sf_close(sndFile);
return 1;
}
// This next step, actually reads in the wav file data. This function automatically converts the whichever format the audio file data is in to doubles. The library can convert to a number of different formats.
// Load data
long numFrames = sf_readf_float(sndFile, buffer, sndInfo.frames);
// Check correct number of samples loaded
if (numFrames != sndInfo.frames) {
fprintf(stderr, "Did not read enough samples for source\n");
sf_close(sndFile);
free(buffer);
return 1;
}
// Now just output some info about the wav file
printf("Read %ld samples from %s, Sample rate: %d, Length: %fs\n",
numFrames, argv[1], sndInfo.samplerate, (float)numFrames/sndInfo.samplerate);
// extract a single channel out
float *buffer_singlechannel = malloc(sndInfo.frames * sizeof(float));
long i=0;
for (long f=0 ; f<numFrames ; f++) {
buffer_singlechannel[f]= buffer[i]; // Channel 1
i+=sndInfo.channels;
}
// free the buffer with multiple channel input
free(buffer);
// array to store the autocorrelation function
double *buffer_autocorrelation = malloc((2*sndInfo.frames +1)* sizeof(double));
// calculate autocorrelation
xcorr( buffer_singlechannel, buffer_singlechannel, buffer_autocorrelation, sndInfo.frames, 0,sndInfo.frames, 1, 1, sndInfo.frames,sndInfo.frames, sndInfo.frames, sndInfo.frames);
// choose delay to be second zero of auto correlation
long long delay = zeroCrossing(buffer_autocorrelation, 2*sndInfo.frames +1, delay_choice);
printf("Delay chosen = %lld\n",delay);
long long delay_embd_length= (sndInfo.frames -(delay*embed_dimension)+2)* embed_dimension;
//create delay embedding and store in a linear array in row major form
float *buffer_delayembedding =malloc(delay_embd_length*sizeof(float));
long long k1=0;
for (long long l=(delay*embed_dimension)-2; l<sndInfo.frames; l=l+1) {
buffer_delayembedding[k1]=buffer_singlechannel[l];
buffer_delayembedding[k1+1]=buffer_singlechannel[l-delay];
buffer_delayembedding[k1+2]=buffer_singlechannel[l-2*delay];
// printf("%f %f %f\n", buffer_delayembedding[k1],buffer_delayembedding[k1+1],buffer_delayembedding[k1+2]);
k1=k1+embed_dimension;
}
int *array =malloc(num_of_points_retained *sizeof(int));
random_selector( delay_embd_length/ embed_dimension ,num_of_points_retained , array);
for (int i=0; i<num_of_points_retained; i++) {
// printf("%d\n",array[i]);
}
float *buffer_delayembedding_selec =malloc(num_of_points_retained*embed_dimension*sizeof(float));
float * density_vals =malloc(num_of_points_retained* sizeof(float));
printf("total len = %lld\n ", delay_embd_length/3 );
printf("buffer len = %ld\n", num_of_points_retained*embed_dimension);
printf("num_of_points_retained = %ld\n", num_of_points_retained);
for (int i=0; i<num_of_points_retained; i=i+1)
{
int loc=(array[i]-1)*embed_dimension;
// printf("i :%d loc=%d array[i]: %d \n",i, loc, array[i]);
buffer_delayembedding_selec[3*i]=buffer_delayembedding[loc];
buffer_delayembedding_selec[3*i+1]=buffer_delayembedding[loc+1];
buffer_delayembedding_selec[3*i+2]=buffer_delayembedding[loc+2];
// printf("buffer selec [%d] = %f ", 3*i, buffer_delayembedding_selec[3*i]);
// printf("[%d] = %f ", 3*i+1, buffer_delayembedding_selec[3*i+1]);
// printf("[%d] = %f \n", 3*i+2, buffer_delayembedding_selec[3*i+2]);
}
// for (int i=0; i<delay_embd_length/ embed_dimension ; i++)
// {
// printf("% d th original :%f %f %f\n",i, buffer_delayembedding [3*i],buffer_delayembedding [3*i+1],buffer_delayembedding [3*i+2]);
// }
knn (buffer_delayembedding_selec, density_vals , (num_of_points_retained), num_neighbours);
// free all arrays used till now except the single channel wave file and close the sndFile object
sf_close(sndFile);
//desnity based subsampling
float * density_vals_temp =malloc(num_of_points_retained* sizeof(float));
memcpy(density_vals_temp,density_vals,num_of_points_retained* sizeof(float));
qsort(density_vals_temp, num_of_points_retained, sizeof(float), cmpfunc);
float key=findNumber(100-percentile,density_vals_temp,500);
// Function call and the print statement
// printf(" \n %d percentile: %.4lf \n", percentile, key);
printf(" \n the linear array output contains the subsampled point cloud \n");
//for (int kk=0; kk<num_of_points_retained; kk++) {
// printf("%d index %f \n", kk,density_vals_temp[kk]);
//}
free(density_vals_temp);
int counter2=0;
for( int counter = 0 ; counter < num_of_points_retained; counter++)
{
if (density_vals[counter]>=key && counter2<output_len/2)
{
output[2*counter2]=buffer_delayembedding_selec[3*counter];
output[2*counter2+1]=buffer_delayembedding_selec[3*counter+1];
//printf("%d output %.4lf , %.4lf\n", counter2,output[2*counter2],output[2*counter2+1]);
//printf("%d output %.4lf , %.4lf\n", counter,buffer_delayembedding_selec[3*counter],buffer_delayembedding_selec[3*counter+1]);
counter2++;
}
}
free(buffer_singlechannel);
free(buffer_autocorrelation);
free(buffer_delayembedding);
free(array);
free(buffer_delayembedding_selec);
free(density_vals);
printf("Density based subsampling End\n");
free(output);
return 0;
}
void vlad_compute(int k, int d, const float *centroids, int n, const float *v,int flags, float *desc)
{
int i,j,l,n_quantile,i0,i1,ai,a,ma,ni;
int *perm ;
float un , diff;
float *tab,*u,*avg,*sum,*mom2,*dists;
int *hist,*assign;
if(flags<11 || flags>=13)
{
assign=ivec_new(n);
nn(n,k,d,centroids,v,assign,NULL,NULL);
if(flags==6 || flags==7)
{
n_quantile = flags==6 ? 3 : 1;
fvec_0(desc,k*d*n_quantile);
perm = ivec_new(n);
tab = fvec_new(n);
ivec_sort_index(assign,n,perm);
i0=0;
for(i=0;i<k;i++)
{
i1=i0;
while(i1<n && assign[perm[i1]]==i)
{
i1++;
}
if(i1==i0) continue;
for(j=0;j<d;j++)
{
for(l=i0;l<i1;l++)
{
tab[l-i0]=v[perm[l]*d+j];
}
ni=i1-i0;
fvec_sort(tab,ni);
for(l=0;l<n_quantile;l++)
{
desc[(i*d+j)*n_quantile+l]=(tab[(l*ni+ni/2)/n_quantile]-centroids[i*d+j])*ni;
}
}
i0=i1;
}
free(perm);
free(tab);
}
else if(flags==5)
{
fvec_0(desc,k*d);
for(i=0;i<n;i++)
{
for(j=0;j<d;j++)
{
desc[assign[i]*d+j]+=v[i*d+j];
}
}
}
else if(flags==8 || flags==9)
{
fvec_0(desc,k*d);
u = fvec_new(d);
for(i=0;i<n;i++)
{
fvec_cpy(u,v+i*d,d);
fvec_sub(u,centroids+assign[i]*d,d);
un=(float)sqrt(fvec_norm2sqr(u,d));
if(un==0) continue;
if(flags==8)
{
fvec_div_by(u,d,un);
} else if(flags==9)
{
fvec_div_by(u,d,sqrt(un));
}
fvec_add(desc+assign[i]*d,u,d);
}
free(u);
}
else if(flags==10)
{
fvec_0(desc,k*d);
for(i=0;i<n;i++)
{
for(j=0;j<d;j++)
{
desc[assign[i]*d+j]+=v[i*d+j];
}
}
for(i=0;i<k;i++)
{
fvec_normalize(desc+i*d,d,2.0);
}
}
else if(flags==13)
{
fvec_0(desc,k*d);
for(i=0;i<n;i++)
{
for(j=0;j<d;j++)
{
desc[assign[i]*d+j]+=(float)sqr(v[i*d+j]-centroids[assign[i]*d+j]);
}
}
}
else if(flags==14)
{
avg = fvec_new_0(k*d);
for(i=0;i<n;i++)
{
for(j=0;j<d;j++)
{
avg[assign[i]*d+j]+=v[i*d+j]-centroids[assign[i]*d+j];
}
}
hist=ivec_new_histogram(k,assign,n);
for(i=0;i<k;i++)
{
if(hist[i]>0)
{
for(j=0;j<d;j++)
{
avg[i*d+j]/=hist[i];
}
}
}
free(hist);
fvec_0(desc,k*d);
for(i=0;i<n;i++)
{
for(j=0;j<d;j++)
{
desc[assign[i]*d+j]+=(float)(sqr(v[i*d+j]-centroids[assign[i]*d+j]-avg[assign[i]*d+j]));
}
}
fvec_sqrt(desc,k*d);
free(avg);
}
else if(flags==15)
{
fvec_0(desc,k*d*2);
sum = desc;
for(i=0;i<n;i++)
{
for(j=0;j<d;j++)
{
sum[assign[i]*d+j]+=v[i*d+j]-centroids[assign[i]*d+j];
}
}
hist = ivec_new_histogram(k,assign,n);
mom2 = desc+k*d;
for(i=0;i<n;i++)
{
ai=assign[i];
for(j=0;j<d;j++)
{
mom2[ai*d+j]+=(float)(sqr(v[i*d+j]-centroids[ai*d+j]-sum[ai*d+j]/hist[ai]));
}
}
fvec_sqrt(mom2,k*d);
free(hist);
}
else if(flags==17)
{
fvec_0(desc,k*d*2);
for(i=0;i<n;i++)
{
for(j=0;j<d;j++)
{
diff=v[i*d+j]-centroids[assign[i]*d+j];
if(diff>0)
{
desc[assign[i]*d+j]+=diff;
}
else
{
desc[assign[i]*d+j+k*d]-=diff;
}
}
}
}
else
{
fvec_0(desc,k*d);
for(i=0;i<n;i++)
{
for(j=0;j<d;j++)
{
desc[assign[i]*d+j]+=v[i*d+j]-centroids[assign[i]*d+j];
}
}
if(flags==1)
{
hist=ivec_new_histogram(k,assign,n);
/* printf("unbalance factor=%g\n",ivec_unbalanced_factor(hist,k)); */
for(i=0;i<k;i++)
{
for(j=0;j<d;j++)
{
desc[i*d+j]/=hist[i];
}
}
free(hist);
}
if(flags==2)
{
for(i=0;i<k;i++)
{
fvec_normalize(desc+i*d,d,2.0);
}
}
if(flags==3 || flags==4)
{
assert(!"not implemented");
}
if(flags==16)
{
hist=ivec_new_histogram(k,assign,n);
for(i=0;i<k;i++)
{
if(hist[i]>0)
{
fvec_norm(desc+i*d,d,2);
fvec_mul_by(desc+i*d,d,sqrt(hist[i]));
}
}
free(hist);
}
}
free(assign);
}
else if(flags==11 || flags==12)
{
ma=flags==11 ? 4 : 2;
assign=ivec_new(n*ma);
dists=knn(n,k,d,ma,centroids,v,assign,NULL,NULL);
fvec_0(desc,k*d);
for(i=0;i<n;i++)
{
for(j=0;j<d;j++)
{
for(a=0;a<ma;a++)
{
desc[assign[ma*i+a]*d+j]+=v[i*d+j]-centroids[assign[ma*i+a]*d+j];
}
}
}
free(dists);
free(assign);
}
}