apaam/mlpack
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Memory leak error for ball-tree has been fixed...

Browse Source
This commit is contained in:
Dongryeol Lee
2008-04-22 19:23:29 +00:00
parent 17f7e9769e
commit 5801a767dc
4 changed files with 59 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastlib2/contrib/dongryel/proximity_project/gen_metric_tree.h
+1
View File
@@ -64,6 +64,7 @@ namespace proximity {
}
node->Init(0, matrix.n_cols());
node->bound().center().Init(matrix.n_rows());
tree_gen_metric_tree_private::SplitGenMetricTree<TMetricTree>
(matrix, node, leaf_size, old_from_new_ptr);
fastlib2/contrib/dongryel/proximity_project/gen_metric_tree_impl.h
+4 -6
View File
@@ -10,7 +10,6 @@ namespace tree_gen_metric_tree_private {
void MakeLeafMetricTreeNode(const Matrix& matrix,
index_t begin, index_t count, TBound *bounds) {
bounds->center().Init(matrix.n_rows());
bounds->center().SetZero();
index_t end = begin + count;
@@ -169,11 +168,11 @@ namespace tree_gen_metric_tree_private {
*left = new TMetricTree();
*right = new TMetricTree();
(*left)->bound().center().Init(matrix.n_rows());
(*right)->bound().center().Init(matrix.n_rows());
((*left)->bound().center()).Init(matrix.n_rows());
((*right)->bound().center()).Init(matrix.n_rows());
(*left)->bound().center().CopyValues(furthest_from_random_row_vec);
(*right)->bound().center().CopyValues
((*left)->bound().center()).CopyValues(furthest_from_random_row_vec);
((*right)->bound().center()).CopyValues
(furthest_from_furthest_random_row_vec);
index_t left_count = MatrixPartition
@@ -192,7 +191,6 @@ namespace tree_gen_metric_tree_private {
TMetricTree *right) {
// First clear the internal node center.
node->bound().center().Init(matrix.n_rows());
node->bound().center().SetZero();
// Compute the weighted sum of the two pivots
fastlib2/contrib/dongryel/proximity_project/main.cc
+1 -1
View File
@@ -28,7 +28,7 @@ int main(int argc, char *argv[]) {
fx_timer_stop(NULL, "pca tree");
printf("Got %d eigenvalues...\n", root_->stat().eigenvalues_.n_cols());
printf("Got %d eigenvalues...\n", root_->stat().eigenvalues_.length());
root_->stat().eigenvalues_.PrintDebug();
printf("Finished constructing the tree...\n");
fastlib2/contrib/dongryel/proximity_project/subspace_stat.h
+53 -23
View File
@@ -5,7 +5,7 @@ class SubspaceStat {
private:
static const double epsilon_ = 0.05;
static const double epsilon_ = 0.1;
void AddVectorToMatrix(Matrix &A, const Vector &v, Matrix &R) {
@@ -91,7 +91,7 @@ class SubspaceStat {
Matrix eigenvectors_;
Matrix eigenvalues_;
Vector eigenvalues_;
/** compute PCA exhaustively for leaf nodes */
void Init(const Matrix& dataset, index_t &start, index_t &count) {
@@ -101,8 +101,8 @@ class SubspaceStat {
Vector point;
dataset.MakeColumnVector(start, &point);
means_.Copy(point);
eigenvalues_.Init(1, 1);
eigenvalues_.set(0, 0, 0);
eigenvalues_.Init(1);
eigenvalues_[0] = 0;
eigenvectors_.Init(dataset.n_rows(), 1);
eigenvectors_.SetZero();
return;
@@ -145,7 +145,7 @@ class SubspaceStat {
}
}
eigenvalues_.Init(eigencount, eigencount);
eigenvalues_.Init(eigencount);
eigenvalues_.SetZero();
eigenvectors_.Init(dataset.n_rows(), eigencount);
@@ -154,9 +154,8 @@ class SubspaceStat {
for(index_t i = 0, index = 0; i < singular_values.length(); i++) {
if(singular_values[i] >= epsilon_ * max_singular_value) {
Vector source, destination;
eigenvalues_.set(index, index,
singular_values[i] * singular_values[i] /
((double) count_));
eigenvalues_[index] = singular_values[i] * singular_values[i] /
((double) count_);
left_singular_vectors.MakeColumnVector(i, &source);
eigenvectors_.MakeColumnVector(index, &destination);
@@ -165,7 +164,12 @@ class SubspaceStat {
}
}
printf("Leaf has %d basis sets...\n", eigenvalues_.n_cols());
/*
printf("Leaf has %d basis sets spanning %d points...\n",
eigenvalues_.length(), count_);
eigenvalues_.PrintDebug();
exit(0);
*/
}
/**
@@ -277,8 +281,8 @@ class SubspaceStat {
const Vector &projection_of_mean_diff,
Matrix *eigensystem) {
Matrix left_eigenbasis, right_eigenbasis,
left_eigenvalues, right_eigenvalues;
Matrix left_eigenbasis, right_eigenbasis;
Vector left_eigenvalues, right_eigenvalues;
// left and right's eigenbasis and the mean vectors
left_eigenbasis.Alias(left_stat.eigenvectors_);
@@ -293,25 +297,35 @@ class SubspaceStat {
((double) count_ * count_);
if(leftside_nullspace_basis.n_cols() > 0) {
eigensystem->Init(left_eigenvalues.n_rows() +
eigensystem->Init(left_eigenvalues.length() +
leftside_nullspace_basis.n_cols(),
left_eigenvalues.n_rows() +
left_eigenvalues.length() +
leftside_nullspace_basis.n_cols());
}
else {
eigensystem->Init(left_eigenvalues.n_rows(), left_eigenvalues.n_rows());
eigensystem->Init(left_eigenvalues.length(), left_eigenvalues.length());
}
eigensystem->SetZero();
// compute the top left part of the eigensystem
Matrix top_left, top_tmp;
la::MulInit(projection_of_right_eigenbasis, right_eigenvalues, &top_tmp);
top_tmp.Init(projection_of_right_eigenbasis.n_rows(),
projection_of_right_eigenbasis.n_cols());
for(index_t i = 0; i < projection_of_right_eigenbasis.n_cols(); i++) {
la::ScaleOverwrite(projection_of_right_eigenbasis.n_rows(),
right_eigenvalues[i],
projection_of_right_eigenbasis.GetColumnPtr(i),
top_tmp.GetColumnPtr(i));
}
//la::MulInit(projection_of_right_eigenbasis, right_eigenvalues, &top_tmp);
la::MulTransBInit(top_tmp, projection_of_right_eigenbasis, &top_left);
for(index_t i = 0; i < left_eigenvalues.n_rows(); i++) {
for(index_t j = 0; j < left_eigenvalues.n_cols(); j++) {
eigensystem->set(i, j, factor1 * left_eigenvalues.get(i, j) +
for(index_t i = 0; i < left_eigenvalues.length(); i++) {
for(index_t j = 0; j < left_eigenvalues.length(); j++) {
double left_eigenvalue_factor = (i == j) ? left_eigenvalues[j]:0;
eigensystem->set(i, j, factor1 * left_eigenvalue_factor +
factor2 * top_left.get(i, j) +
factor3 * projection_of_mean_diff[i] *
projection_of_mean_diff[j]);
@@ -346,8 +360,21 @@ class SubspaceStat {
}
}
la::MulInit(proj_rightside_eigenbasis_on_leftside_nullspace,
right_eigenvalues, &bottom_tmp);
bottom_tmp.Init
(proj_rightside_eigenbasis_on_leftside_nullspace.n_rows(),
proj_rightside_eigenbasis_on_leftside_nullspace.n_cols());
for(index_t i = 0;
i < proj_rightside_eigenbasis_on_leftside_nullspace.n_cols(); i++) {
la::ScaleOverwrite
(proj_rightside_eigenbasis_on_leftside_nullspace.n_rows(),
right_eigenvalues[i],
proj_rightside_eigenbasis_on_leftside_nullspace.GetColumnPtr(i),
bottom_tmp.GetColumnPtr(i));
}
//la::MulInit(proj_rightside_eigenbasis_on_leftside_nullspace,
// right_eigenvalues, &bottom_tmp);
la::MulTransBInit(bottom_tmp, projection_of_right_eigenbasis,
&bottom_left);
@@ -447,7 +474,7 @@ class SubspaceStat {
eigencount++;
}
}
eigenvalues_.Init(eigencount, eigencount);
eigenvalues_.Init(eigencount);
eigenvalues_.SetZero();
// relationship between the singular value and the eigenvalue is
@@ -457,7 +484,7 @@ class SubspaceStat {
for(index_t i = 0, index = 0; i < evalues.length(); i++) {
if(evalues[i] >= epsilon_ * max_eigenvalue) {
Vector s, d;
eigenvalues_.set(index, index, evalues[i]);
eigenvalues_[index] = evalues[i];
eigenvectors_.MakeColumnVector(i, &s);
tmp_eigenvectors.MakeColumnVector(index, &d);
d.CopyValues(s);
@@ -474,7 +501,10 @@ class SubspaceStat {
la::Scale(factor1, &means_);
la::AddExpert(factor2, right_stat.means_, &means_);
printf("Internal has %d basis sets...\n", eigenvectors_.n_cols());
/*
printf("Internal has %d basis sets spanning %d points...\n",
eigenvectors_.n_cols(), count_);
*/
}
SubspaceStat() { }