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

Undid a horrible mistake

Browse Source
This commit is contained in:
gcolon7
2010-11-18 19:00:57 +00:00
parent 7e28755598
commit ca61cfa603
16 changed files with 89 additions and 73 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastlib/branches/fastlib-stl/contrib/nvasil/convex_nmf/gop_nmf_impl.h
+2 -2
View File
@@ -706,7 +706,7 @@ bool RelaxedRescaledNmfL1::IsIntermediateStepOver(Matrix &coordinates,
double norm_gradient=la::Dot(gradient.n_elements(),
gradient.ptr(),
gradient.ptr());
if (step/std::pow(2, norm_gradient) < grad_tolerance_) {
if (math::Pow<1,2>(norm_gradient)*step < grad_tolerance_) {
return true;
}
return false;
@@ -1269,7 +1269,7 @@ bool RelaxedNmfIsometric::IsOptimizationOver(Matrix &coordinates,
bool RelaxedNmfIsometric::IsIntermediateStepOver(Matrix &coordinates,
Matrix &gradient,
double step) {
double norm_gradient=1.0/std::pow(2,la::Dot(gradient.n_elements(),
double norm_gradient=math::Pow<1,2>(la::Dot(gradient.n_elements(),
gradient.ptr(),
gradient.ptr()));
if (norm_gradient*step < grad_tolerance_) {
fastlib/branches/fastlib-stl/contrib/nvasil/convex_nmf/sdp_objectives_impl.h
+8 -8
View File
@@ -87,7 +87,7 @@ void SmallSdpNmf::ComputeGradient(Matrix &coordinates, Matrix *gradient) {
//double w=coordinates.get(j, w_i);
double t1=coordinates.get(j, t1_i);
double dw=-1.0;
double dt1=0.5/std::pow(-2,t1);
double dt1=0.5*math::Pow<-1,2>(t1);
gradient->set(j, w_i, dw);
gradient->set(j, t1_i, dt1);
}
@@ -100,7 +100,7 @@ void SmallSdpNmf::ComputeGradient(Matrix &coordinates, Matrix *gradient) {
// double h=coordinates.get(j, h_i);
double t2=coordinates.get(j, t2_i);
double dh=-1.0;
double dt2=0.5/std::pow(-2,t2);
double dt2=0.5*math::Pow<-1,2>(t2);
gradient->set(j, h_i, dh);
gradient->set(j, t2_i, dt2);
}
@@ -134,7 +134,7 @@ void SmallSdpNmf::ComputeGradient(Matrix &coordinates, Matrix *gradient) {
}
}
// from the SDP cones
// determinant=(t1-w*w)*(t2-h*h)-std::pow(2,w*h-v);
// determinant=(t1-w*w)*(t2-h*h)-math::Pow<2,1>(w*h-v);
// dw=-2*w*(t2-h*h)-2*h*(w*h-v);
// dh=-2*h*(t1-w*w)-2*w*(w*h-v);
// dt1=t2-h*h;
@@ -156,7 +156,7 @@ void SmallSdpNmf::ComputeGradient(Matrix &coordinates, Matrix *gradient) {
double t2_minus_hh=(t2-h*h);
double wh_minus_v=(w*h-v);
double determinant=t1_minus_ww*t2_minus_hh
-std::pow(2,wh_minus_v);
-math::Pow<2,1>(wh_minus_v);
//NOTIFY("determinant:%lg", determinant);
if (determinant<=0.0) {
@@ -196,7 +196,7 @@ void SmallSdpNmf::ComputeObjective(Matrix &coordinates, double *objective) {
for(index_t j=0; j<new_dim_; j++) {
double w=coordinates.get(j, w_i);
double t1=coordinates.get(j, t1_i);
*objective+=1/std::pow(2,t1)-w;
*objective+=math::Pow<1,2>(t1)-w;
}
}
for(index_t i=0; i<num_of_columns_; i++) {
@@ -205,7 +205,7 @@ void SmallSdpNmf::ComputeObjective(Matrix &coordinates, double *objective) {
for(index_t j=0; j<new_dim_; j++) {
double h=coordinates.get(j, h_i);
double t2=coordinates.get(j, t2_i);
*objective+=1/std::pow(2,t2)-h;
*objective+=math::Pow<1,2>(t2)-h;
}
}
//Vector temp;
@@ -250,7 +250,7 @@ double SmallSdpNmf::ComputeLagrangian(Matrix &coordinates) {
}
lagrangian-=log(temp_prod);
// from the SDP cones
// determinant=(t1-w*w)*(t2-h*h)-std::pow(2,w*h-v);
// determinant=(t1-w*w)*(t2-h*h)-math::Pow<2,1>(w*h-v);
temp_prod=1.0;
for(index_t i=0; i<values_.size(); i++) {
index_t w_i=rows_[i];
@@ -275,7 +275,7 @@ double SmallSdpNmf::ComputeLagrangian(Matrix &coordinates) {
}
double wh_minus_v=(w*h-v);
double determinant=t1_minus_ww*t2_minus_hh
-std::pow(2,wh_minus_v);
-math::Pow<2,1>(wh_minus_v);
if (unlikely(determinant<=0)) {
return DBL_MAX;
}
fastlib/branches/fastlib-stl/contrib/nvasil/mvu/mvu_classification_impl.h
+8 -8
View File
@@ -81,7 +81,7 @@ void MaxFurthestNeighborsSemiSupervised::Init(fx_module *module, Matrix &labeled
}
sum_of_nearest_distances_=0;
for(index_t i=0; i<nearest_distances_.size(); i++) {
sum_of_nearest_distances_+=1.0/std::pow(2,nearest_distances_[i]);
sum_of_nearest_distances_+=math::Pow<1,2>(nearest_distances_[i]);
}
NOTIFY("Sum of all nearest distances:%lg", sum_of_nearest_distances_);
fx_result_double(module_, "sum_of_nearest_distances", sum_of_nearest_distances_);
@@ -323,7 +323,7 @@ bool MaxFurthestNeighborsSemiSupervised::IsOptimizationOver(
bool MaxFurthestNeighborsSemiSupervised::IsIntermediateStepOver(
Matrix &coordinates, Matrix &gradient, double step) {
double norm_gradient=1.0/std::pow(2,la::Dot(gradient.n_elements(),
double norm_gradient=math::Pow<1,2>(la::Dot(gradient.n_elements(),
gradient.ptr(),
gradient.ptr()));
double feasibility_error;
@@ -445,7 +445,7 @@ void MaxFurthestNeighborsSvmSemiSupervised::Init(fx_module *module,
}
sum_of_nearest_distances_=0;
for(index_t i=0; i<nearest_distances_.size(); i++) {
sum_of_nearest_distances_+=1.0/std::pow(2,nearest_distances_[i]);
sum_of_nearest_distances_+=math::Pow<1,2>(nearest_distances_[i]);
}
NOTIFY("Sum of all nearest distances:%lg", sum_of_nearest_distances_);
fx_result_double(module_, "sum_of_nearest_distances", sum_of_nearest_distances_);
@@ -718,7 +718,7 @@ double MaxFurthestNeighborsSvmSemiSupervised::ComputeLagrangian(Matrix &coordina
if (sigma1_ * ineq <= ineq_lagrange_mult_[i*num_of_labeled_+j]) {
lagrangian+=(-ineq_lagrange_mult_[i*num_of_labeled_+j] + sigma1_/2*ineq)*ineq;
} else {
lagrangian+=-std::pow(2,ineq_lagrange_mult_[i*num_of_labeled_+j])/(2*sigma1_);
lagrangian+=-math::Pow<2,1>(ineq_lagrange_mult_[i*num_of_labeled_+j])/(2*sigma1_);
}
}
}
@@ -788,7 +788,7 @@ bool MaxFurthestNeighborsSvmSemiSupervised::IsOptimizationOver(
bool MaxFurthestNeighborsSvmSemiSupervised::IsIntermediateStepOver(
Matrix &coordinates, Matrix &gradient, double step) {
double norm_gradient=1.0/std::pow(2,la::Dot(gradient.n_elements(),
double norm_gradient=math::Pow<1,2>(la::Dot(gradient.n_elements(),
gradient.ptr(),
gradient.ptr()));
double feasibility_error;
@@ -912,7 +912,7 @@ void MaxFurthestNeighborsSvmSemiSupervised1::Init(fx_module *module,
}
sum_of_nearest_distances_=0;
for(index_t i=0; i<nearest_distances_.size(); i++) {
sum_of_nearest_distances_+=1.0/std::pow(2,nearest_distances_[i]);
sum_of_nearest_distances_+=math::Pow<1,2>(nearest_distances_[i]);
}
NOTIFY("Sum of all nearest distances:%lg", sum_of_nearest_distances_);
fx_result_double(module_, "sum_of_nearest_distances", sum_of_nearest_distances_);
@@ -1185,7 +1185,7 @@ double MaxFurthestNeighborsSvmSemiSupervised1::ComputeLagrangian(Matrix &coordin
if (sigma1_ * ineq <= ineq_lagrange_mult_[i*num_of_labeled_+j]) {
lagrangian+=(-ineq_lagrange_mult_[i*num_of_labeled_+j] + sigma1_/2*ineq)*ineq;
} else {
lagrangian+=-std::pow(2,ineq_lagrange_mult_[i*num_of_labeled_+j])/(2*sigma1_);
lagrangian+=-math::Pow<2,1>(ineq_lagrange_mult_[i*num_of_labeled_+j])/(2*sigma1_);
}
}
}
@@ -1255,7 +1255,7 @@ bool MaxFurthestNeighborsSvmSemiSupervised1::IsOptimizationOver(
bool MaxFurthestNeighborsSvmSemiSupervised1::IsIntermediateStepOver(
Matrix &coordinates, Matrix &gradient, double step) {
double norm_gradient=1.0/std::pow(2,la::Dot(gradient.n_elements(),
double norm_gradient=math::Pow<1,2>(la::Dot(gradient.n_elements(),
gradient.ptr(),
gradient.ptr()));
double feasibility_error;
fastlib/branches/fastlib-stl/contrib/nvasil/mvu/mvu_objectives_impl.h
+5 -5
View File
@@ -564,7 +564,7 @@ void MaxFurthestNeighbors::Init(datanode *module, Matrix &data) {
&nearest_distances_,
&num_of_nearest_pairs_);
}
sum_of_nearest_distances_=1.0/std::pow(2,la::Dot(nearest_distances_.size(),
sum_of_nearest_distances_=math::Pow<1,2>(la::Dot(nearest_distances_.size(),
&nearest_distances_[0], &nearest_distances_[0]));
fx_format_result(module_, "num_of_constraints", "%i", num_of_nearest_pairs_);
eq_lagrange_mult_.Init(num_of_nearest_pairs_);
@@ -633,7 +633,7 @@ void MaxFurthestNeighbors::Init(fx_module *module) {
}
num_of_points_++;
num_of_nearest_pairs_=nearest_neighbor_pairs_.size();
sum_of_nearest_distances_=1.0/std::pow(2,la::Dot(nearest_distances_.size(),
sum_of_nearest_distances_=math::Pow<1,2>(la::Dot(nearest_distances_.size(),
&nearest_distances_[0], &nearest_distances_[0]));
fclose(fp);
@@ -743,7 +743,7 @@ void MaxFurthestNeighbors::ComputeFeasibilityError(Matrix &coordinates, double *
-nearest_distances_[i];
*error+=dist_diff*dist_diff;
}
*error= 100.0 / std::pow(2,*error)/sum_of_nearest_distances_;
*error= 100 * math::Pow<1,2>(*error)/sum_of_nearest_distances_;
}
double MaxFurthestNeighbors::ComputeLagrangian(Matrix &coordinates) {
@@ -825,7 +825,7 @@ bool MaxFurthestNeighbors::IsOptimizationOver(Matrix &coordinates,
bool MaxFurthestNeighbors::IsIntermediateStepOver(Matrix &coordinates,
Matrix &gradient, double step) {
double norm_gradient=1/std::pow(2,la::Dot(gradient.n_elements(),
double norm_gradient=math::Pow<1,2>(la::Dot(gradient.n_elements(),
gradient.ptr(),
gradient.ptr()));
double feasibility_error;
@@ -894,7 +894,7 @@ void MaxVarianceUtils::EstimateKnns(ArrayList<index_t> &neares_neighbors,
double probability=0;
for(index_t j=0; j<k; j++) {
probability+=exp(-nearest_distances[i*maximum_knns+j]
*std::pow(2,nearest_distances[i*maximum_knns+k]-1))/scale_factor;
/(2*math::Pow<1,2>(nearest_distances[i*maximum_knns+k])))/scale_factor;
}
loocv_score+=log(probability);
mean_band+=nearest_distances[i*maximum_knns+k];
fastlib/branches/fastlib-stl/contrib/nvasil/nmf_tree/relaxed_nmf_bound_tightener_impl.h
+2 -2
View File
@@ -665,7 +665,7 @@ bool RelaxedNmfIsometricBoundTightener::IsOptimizationOver(Matrix &coordinates,
bool RelaxedNmfIsometricBoundTightener::IsIntermediateStepOver(Matrix &coordinates,
Matrix &gradient,
double step) {
double norm_gradient=1.0/std::pow(2,la::Dot(gradient.n_elements(),
double norm_gradient=math::Pow<1,2>(la::Dot(gradient.n_elements(),
gradient.ptr(),
gradient.ptr()));
if (norm_gradient*step < grad_tolerance_) {
@@ -1055,7 +1055,7 @@ bool RelaxedNmfIsometricBoxTightener::IsOptimizationOver(Matrix &coordinates,
bool RelaxedNmfIsometricBoxTightener::IsIntermediateStepOver(Matrix &coordinates,
Matrix &gradient,
double step) {
double norm_gradient=1.0/std::pow(2,la::Dot(gradient.n_elements(),
double norm_gradient=math::Pow<1,2>(la::Dot(gradient.n_elements(),
gradient.ptr(),
gradient.ptr()));
if (norm_gradient*step < grad_tolerance_) {
fastlib/branches/fastlib-stl/contrib/nvasil/non_convex_mvu/non_convex_mvu_impl.h
+1 -1
View File
@@ -134,7 +134,7 @@ void NonConvexMVU::ComputeLocalOptimumSGD() {
index_t exterior_points=0;
for(index_t kk=0; kk<3; kk++) {
for(it1=0; it1<100; it1++) {
step=1.0/std::pow(2,step0/(it1+1));
step=math::Pow<1,2>(step0/(it1+1));
for(index_t it2=0; it2<max_iterations_; it2++) {
for(index_t i=0; i<num_of_points_; i++) {
for(index_t k=0; k<kfns_; k++) {
fastlib/branches/fastlib-stl/contrib/nvasil/non_convex_nmf/main.cc
+4 -4
View File
@@ -116,7 +116,7 @@ void ComputeSpectrum(Matrix &w_mat, Matrix &h_mat, Matrix *spectrum) {
double total_sum=0;
for(index_t i=0; i<w_mat.n_rows(); i++) {
for(index_t j=0; j<w_mat.n_cols(); j++) {
spectrum->set(i, 0, spectrum->get(i, 0)+std::pow(2,w_mat.get(i, j)));
spectrum->set(i, 0, spectrum->get(i, 0)+math::Pow<2,1>(w_mat.get(i, j)));
}
}
Matrix h_norms;
@@ -128,7 +128,7 @@ void ComputeSpectrum(Matrix &w_mat, Matrix &h_mat, Matrix *spectrum) {
}
}
for(index_t i=0; i<h_norms.n_rows(); i++) {
spectrum->set(i, 0, spectrum->get(i, 0)*std::pow(2,h_norms.get(i,0)));
spectrum->set(i, 0, spectrum->get(i, 0)/math::Pow<1,2>(h_norms.get(i,0)));
}
std::sort(spectrum->ptr(),
spectrum->ptr()+spectrum->n_rows(),
@@ -145,12 +145,12 @@ void ComputeNeighborhoodErrors(ArrayList<std::pair<index_t, index_t> > &neighbor
for(index_t i=0; i<neighbor_pairs.size(); i++) {
index_t n1=neighbor_pairs[i].first;
index_t n2=neighbor_pairs[i].second;
*error += std::pow(2,la::DistanceSqEuclidean(new_points.n_rows(),
*error += math::Pow<2,1>(la::DistanceSqEuclidean(new_points.n_rows(),
new_points.GetColumnPtr(n1),
new_points.GetColumnPtr(n2))
-distances[i]);
total_distances += distances[i]*distances[i];
}
*error=100.0 / std::pow(2,*error/total_distances);
*error=100.0 * math::Pow<1,2>(*error/total_distances);
}
fastlib/branches/fastlib-stl/contrib/nvasil/non_convex_nmf/nmf_objectives_impl.h
+16 -16
View File
@@ -310,7 +310,7 @@ double BigSdpNmfObjectiveMinVarIneq::ComputeLagrangian(Matrix &coordinates) {
if (sigma_*diff <= ineq_lagrange_mult_[i]) {
lagrangian+=(sigma_*diff-ineq_lagrange_mult_[i])*diff;
} else {
lagrangian+=-std::pow(2,ineq_lagrange_mult_[i])/sigma_;
lagrangian+=-math::Pow<2,1>(ineq_lagrange_mult_[i])/sigma_;
}
}
return lagrangian;
@@ -529,7 +529,7 @@ double BigSdpNmfObjectiveMinVarDiagonalDominance::ComputeLagrangian(Matrix &coor
if (sigma2_*diff <= ineq_lagrange_mult_[i]) {
lagrangian+=(sigma2_*diff/2-ineq_lagrange_mult_[i])*diff;
} else {
lagrangian+=-std::pow(2,ineq_lagrange_mult_[i])/(2*sigma2_);
lagrangian+=-math::Pow<2,1>(ineq_lagrange_mult_[i])/(2*sigma2_);
}
}
@@ -540,7 +540,7 @@ double BigSdpNmfObjectiveMinVarDiagonalDominance::ComputeLagrangian(Matrix &coor
if (sigma2_*diff < ineq_lagrange_mult1_[i*new_dim_+j]) {
lagrangian+=(sigma2_*diff-ineq_lagrange_mult1_[i*new_dim_+j])*diff;
} else {
lagrangian+=-std::pow(2,
lagrangian+=-math::Pow<2,1>(
ineq_lagrange_mult1_[i*new_dim_+j])/(2*sigma2_);
}
}
@@ -807,8 +807,8 @@ void BigSdpNmfObjectiveMaxVarIsometric::ComputeFeasibilityError(Matrix &coordina
coordinates.GetColumnPtr(h))-values_[i];
*error+=diff*diff;
}
infeasibility1_=100.0/std::pow(2,*error/v_norm_);
printf("dot_infeasibility:%lg%% ", 100.0/std::pow(2,*error/v_norm_)); ;
infeasibility1_=math::Pow<1,2>(*error/v_norm_)*100;
printf("dot_infeasibility:%lg%% ", math::Pow<1,2>(*error/v_norm_)*100); ;
// local isometry
double error2=0;
for(index_t i=0; i<num_of_nearest_pairs_; i++) {
@@ -824,8 +824,8 @@ void BigSdpNmfObjectiveMaxVarIsometric::ComputeFeasibilityError(Matrix &coordina
-nearest_distances_[i]);
}
infeasibility2_=100.0/std::pow(2,error2/sum_all_distances_);
printf("dist_infeasibility:%lg %%\n", 100.0/std::pow(2,error2/sum_all_distances_));
infeasibility2_=math::Pow<1,2>(error2/sum_all_distances_)*100;
printf("dist_infeasibility:%lg %%\n", math::Pow<1,2>(error2/sum_all_distances_)*100);
printf("sigma_ratio:%lg\n", sigma_ratio_);
// sigma_ratio_=infeasibility2_/infeasibility1_;
}
@@ -939,7 +939,7 @@ bool BigSdpNmfObjectiveMaxVarIsometric::IsOptimizationOver(
bool BigSdpNmfObjectiveMaxVarIsometric::IsIntermediateStepOver(
Matrix &coordinates, Matrix &gradient, double step) {
double norm_gradient=1.0/(2,la::Dot(gradient.n_elements(),
double norm_gradient=math::Pow<1,2>(la::Dot(gradient.n_elements(),
gradient.ptr(),
gradient.ptr()));
double feasibility_error;
@@ -1166,7 +1166,7 @@ void BigSdpNmfObjectiveMaxVarIsometricRemoveMean::ComputeFeasibilityError(Matrix
coordinates.GetColumnPtr(h))-values_[i];
*error+=diff*diff;
}
infeasibility1_=1.0/std::pow(2,*error/v_norm_)*100;
infeasibility1_=math::Pow<1,2>(*error/v_norm_)*100;
printf("dot_infeasibility:%lg%% ", infeasibility1_); ;
// local isometry
double error2=0;
@@ -1183,7 +1183,7 @@ void BigSdpNmfObjectiveMaxVarIsometricRemoveMean::ComputeFeasibilityError(Matrix
-nearest_distances_[i]);
}
infeasibility2_=1.0/std::pow(2,error2/sum_all_distances_)*100;
infeasibility2_=math::Pow<1,2>(error2/sum_all_distances_)*100;
printf("dist_infeasibility:%lg %%\n", infeasibility2_);
}
@@ -1291,7 +1291,7 @@ bool BigSdpNmfObjectiveMaxVarIsometricRemoveMean::IsOptimizationOver(
bool BigSdpNmfObjectiveMaxVarIsometricRemoveMean::IsIntermediateStepOver(
Matrix &coordinates, Matrix &gradient, double step) {
double norm_gradient=1.0/std::pow(2,la::Dot(gradient.n_elements(),
double norm_gradient=math::Pow<1,2>(la::Dot(gradient.n_elements(),
gradient.ptr(),
gradient.ptr()));
double feasibility_error;
@@ -1524,7 +1524,7 @@ void BigSdpNmfObjectiveMaxFurthestIsometric::ComputeFeasibilityError(Matrix &coo
coordinates.GetColumnPtr(h))-values_[i];
*error+=diff*diff;
}
infeasibility1_=100.0/std::pow(2,*error/v_norm_);
infeasibility1_=math::Pow<1,2>(*error/v_norm_)*100;
// local isometry
double error2=0;
for(index_t i=0; i<num_of_nearest_pairs_; i++) {
@@ -1540,7 +1540,7 @@ void BigSdpNmfObjectiveMaxFurthestIsometric::ComputeFeasibilityError(Matrix &coo
-nearest_distances_[i]);
}
infeasibility2_=100.0/std::pow(2,error2/sum_all_distances_);
infeasibility2_=math::Pow<1,2>(error2/sum_all_distances_)*100;
NOTIFY("dot_infeasibility:%lg%% dist_infeasibility:%lg %%",
infeasibility1_, infeasibility2_);
// NOTIFY("sigma_ratio:%lg\n", sigma_ratio_);
@@ -1656,7 +1656,7 @@ bool BigSdpNmfObjectiveMaxFurthestIsometric::IsOptimizationOver(
bool BigSdpNmfObjectiveMaxFurthestIsometric::IsIntermediateStepOver(
Matrix &coordinates, Matrix &gradient, double step) {
double norm_gradient=1.0/std::pow(2,la::Dot(gradient.n_elements(),
double norm_gradient=math::Pow<1,2>(la::Dot(gradient.n_elements(),
gradient.ptr(),
gradient.ptr()));
double feasibility_error;
@@ -1856,7 +1856,7 @@ void NmfObjectiveIsometric::ComputeFeasibilityError(Matrix &coordinates,
-nearest_distances_[i]);
}
infeasibility2_=100.0/std::pow(2,*error/sum_all_distances_);
infeasibility2_=math::Pow<1,2>(*error/sum_all_distances_)*100;
printf("dist_infeasibility:%lg %%\n", infeasibility2_*100);
}
@@ -1933,7 +1933,7 @@ bool NmfObjectiveIsometric::IsOptimizationOver(
bool NmfObjectiveIsometric::IsIntermediateStepOver(
Matrix &coordinates, Matrix &gradient, double step) {
double norm_gradient=1.0/std::pow(2,la::Dot(gradient.n_elements(),
double norm_gradient=math::Pow<1,2>(la::Dot(gradient.n_elements(),
gradient.ptr(),
gradient.ptr()));
double feasibility_error;
fastlib/branches/fastlib-stl/fastlib/fastlib.h
-1
View File
@@ -38,7 +38,6 @@
#ifndef FASTLIB_FASTLIB_H
#define FASTLIB_FASTLIB_H
#include <cmath>
#include "base/base.h"
#include "la/la.h"
#include "la/matrix.h"
fastlib/branches/fastlib-stl/fastlib/la/la.h
+2 -2
View File
@@ -91,7 +91,7 @@ namespace la {
double s = 0;
do {
double d = *va++ - *vb++;
s += std::pow(t_pow,d);
s += math::PowAbs<t_pow, 1>(d);
} while (--length);
return s;
}
@@ -107,7 +107,7 @@ namespace la {
template<int t_pow>
inline double LMetric(
index_t length, const double *va, const double *vb) {
return 1.0/std::pow(t_pow, RawLMetric<t_pow>(length, va, vb));
return math::Pow<1, t_pow>(RawLMetric<t_pow>(length, va, vb));
}
/** Finds the trace of the matrix.
* Trace(A) is the sum of the diagonal elements
fastlib/branches/fastlib-stl/fastlib/optimization/lbfgs/lbfgs_impl.h
+1 -1
View File
@@ -620,7 +620,7 @@ std::string Lbfgs<OptimizedFunction>::ComputeProgress_() {
double feasibility_error;
optimized_function_->ComputeFeasibilityError(
coordinates_, &feasibility_error);
double norm_grad = 1.0 / std::pow(2, la::Dot(gradient_.n_elements(),
double norm_grad=math::Pow<1,2>(la::Dot(gradient_.n_elements(),
gradient_.ptr(), gradient_.ptr()));
char buffer[1024];
sprintf(buffer, "iteration:%i sigma:%lg lagrangian:%lg objective:%lg error:%lg "
fastlib/branches/fastlib-stl/fastlib/optimization/lbfgs/optimization_utils.h
+3 -3
View File
@@ -170,8 +170,8 @@ class OptUtils {
ones.SetAll(1.0);
// This part of the sparsity constraint function formula can be
// precomputed and it is the same for every iteration
double precomputed_sparse_factor=-sparse_factor*(1.0/std::pow(2,dimension)-1)+
1.0/std::pow(2,dimension);
double precomputed_sparse_factor=-sparse_factor*(math::Pow<1,2>(dimension)-1)+
math::Pow<1,2>(dimension);
for (index_t i=0; i<data->n_cols(); i++) {
double *point=data->GetColumnPtr(i);
@@ -199,7 +199,7 @@ class OptUtils {
double w_norm=la::LengthEuclidean(w_vector);
double w_times_v = 2*la::Dot(v_vector, w_vector);
double v_norm_minus_l2=la::LengthEuclidean(v_vector)-l2_norm;
double alpha = (-w_times_v+1.0/std::pow(2,w_times_v*w_times_v
double alpha = (-w_times_v+math::Pow<1,2>(w_times_v*w_times_v
-4*w_norm*v_norm_minus_l2))/(2*w_norm);
la::AddExpert(alpha, w_vector, &v_vector);
bool all_positive=true;
fastlib/branches/fastlib-stl/fastlib/tree/dballbound_impl.h
+9 -9
View File
@@ -76,7 +76,7 @@ double DBallBound<TMetric, TPoint>::MinDistance(const Point& point) const {
template<typename TMetric, typename TPoint>
double DBallBound<TMetric, TPoint>::MinDistanceSq(const Point& point) const {
return std::pow(2,MinDistance(point));
return math::Pow<2, 1>(MinDistance(point));
}
/**
@@ -90,7 +90,7 @@ double DBallBound<TMetric, TPoint>::MinDistance(const DBallBound& other) const {
template<typename TMetric, typename TPoint>
double DBallBound<TMetric, TPoint>::MinDistanceSq(const DBallBound& other) const {
return std::pow(2,MinDistance(other));
return math::Pow<2, 1>(MinDistance(other));
}
/**
@@ -103,7 +103,7 @@ double DBallBound<TMetric, TPoint>::MaxDistance(const Point& point) const {
template<typename TMetric, typename TPoint>
double DBallBound<TMetric, TPoint>::MaxDistanceSq(const Point& point) const {
return std::pow(2,MaxDistance(point));
return math::Pow<2, 1>(MaxDistance(point));
}
/**
@@ -116,7 +116,7 @@ double DBallBound<TMetric, TPoint>::MaxDistance(const DBallBound& other) const {
template<typename TMetric, typename TPoint>
double DBallBound<TMetric, TPoint>::MaxDistanceSq(const DBallBound& other) const {
return std::pow(2, MaxDistance(other));
return math::Pow<2, 1>(MaxDistance(other));
}
/**
@@ -138,8 +138,8 @@ DRange DBallBound<TMetric, TPoint>::RangeDistanceSq(const DBallBound& other) con
double delta = MidDistance(other.center_);
double sumradius = radius_ + other.radius_;
return DRange(
std::pow(2,math::ClampNonNegative(delta - sumradius)),
std::pow(2,delta + sumradius));
math::Pow<2, 1>(math::ClampNonNegative(delta - sumradius)),
math::Pow<2, 1>(delta + sumradius));
}
/**
@@ -161,7 +161,7 @@ double DBallBound<TMetric, TPoint>::MinToMid(const DBallBound& other) const {
template<typename TMetric, typename TPoint>
double DBallBound<TMetric, TPoint>::MinToMidSq(const DBallBound& other) const {
return std::pow(2, MinToMid(other));
return math::Pow<2, 1>(MinToMid(other));
}
/**
@@ -175,7 +175,7 @@ double DBallBound<TMetric, TPoint>::MinimaxDistance(const DBallBound& other) con
template<typename TMetric, typename TPoint>
double DBallBound<TMetric, TPoint>::MinimaxDistanceSq(const DBallBound& other) const {
return std::pow(2, MinimaxDistance(other));
return math::Pow<2, 1>(MinimaxDistance(other));
}
/**
@@ -197,7 +197,7 @@ double DBallBound<TMetric, TPoint>::MidDistance(const DBallBound& other) const {
template<typename TMetric, typename TPoint>
double DBallBound<TMetric, TPoint>::MidDistanceSq(const DBallBound& other) const {
return std::pow(2, MidDistance(other));
return math::Pow<2, 1>(MidDistance(other));
}
template<typename TMetric, typename TPoint>
fastlib/branches/fastlib-stl/fastlib/tree/dhrectbound_impl.h
+17
View File
@@ -271,6 +271,23 @@ double DHrectBound<t_pow>::MaxDistanceSq(const vec& point) const {
return pow(sum, 2.0 / (double) t_pow);
}
/**
* Calculates maximum bound-to-point squared distance.
*/
/*
template<int t_pow>
double DHrectBound<t_pow>::MaxDistanceSq(const double *point) const {
double sum = 0;
for (index_t d = 0; d < dim_; d++) {
double v = std::max(point[d] - bounds_[d].lo, bounds_[d].hi - point[d]);
sum += math::Pow<t_pow, 1>(v); // v is non-negative
}
return math::Pow<2, t_pow>(sum);
}
*/
/**
* Computes maximum distance.
*/
fastlib/branches/fastlib-stl/mlpack/mvu/mvu_objectives_impl.h
+5 -5
View File
@@ -302,7 +302,7 @@ void MaxFurthestNeighbors::Init(datanode *module, Matrix &data) {
&nearest_distances_,
&num_of_nearest_pairs_);
}
sum_of_nearest_distances_= 1.0/std::pow(2,la::Dot(nearest_distances_.size(),
sum_of_nearest_distances_=math::Pow<1,2>(la::Dot(nearest_distances_.size(),
&nearest_distances_[0], &nearest_distances_[0]));
fx_format_result(module_, "num_of_constraints", "%i", num_of_nearest_pairs_);
eq_lagrange_mult_.Init(num_of_nearest_pairs_);
@@ -371,7 +371,7 @@ void MaxFurthestNeighbors::Init(fx_module *module) {
}
num_of_points_++;
num_of_nearest_pairs_=nearest_neighbor_pairs_.size();
sum_of_nearest_distances_= 1.0/std::pow(2,la::Dot(nearest_distances_.size(),
sum_of_nearest_distances_=math::Pow<1,2>(la::Dot(nearest_distances_.size(),
&nearest_distances_[0], &nearest_distances_[0]));
fclose(fp);
@@ -481,7 +481,7 @@ void MaxFurthestNeighbors::ComputeFeasibilityError(Matrix &coordinates, double *
-nearest_distances_[i];
*error+=dist_diff*dist_diff;
}
*error= 100 / std::pow(2,*error)/sum_of_nearest_distances_;
*error= 100 * math::Pow<1,2>(*error)/sum_of_nearest_distances_;
}
double MaxFurthestNeighbors::ComputeLagrangian(Matrix &coordinates) {
@@ -563,7 +563,7 @@ bool MaxFurthestNeighbors::IsOptimizationOver(Matrix &coordinates,
bool MaxFurthestNeighbors::IsIntermediateStepOver(Matrix &coordinates,
Matrix &gradient, double step) {
double norm_gradient=1.0/std::pow(2,la::Dot(gradient.n_elements(),
double norm_gradient=math::Pow<1,2>(la::Dot(gradient.n_elements(),
gradient.ptr(),
gradient.ptr()));
double feasibility_error;
@@ -632,7 +632,7 @@ void MaxVarianceUtils::EstimateKnns(ArrayList<index_t> &neares_neighbors,
double probability=0;
for(index_t j=0; j<k; j++) {
probability+=exp(-nearest_distances[i*maximum_knns+j]
*std::pow(2,nearest_distances[i*maximum_knns+k]-1))/scale_factor;
/(2*math::Pow<1,2>(nearest_distances[i*maximum_knns+k])))/scale_factor;
}
loocv_score+=log(probability);
mean_band+=nearest_distances[i*maximum_knns+k];
fastlib/branches/fastlib-stl/mlpack/series_expansion/bounds_aux.h
+6 -6
View File
@@ -32,9 +32,9 @@ class bounds_aux {
furthest_point_in_bound1[d] = bound1_range.hi;
v = v2;
}
furthest_dsqd += std::pow(std::fabs(t_pow),v); // v is non-negative
furthest_dsqd += math::PowAbs<t_pow, 1>(v); // v is non-negative
}
furthest_dsqd = std::pow(2.0/t_pow, furthest_dsqd);
furthest_dsqd = math::Pow<2, t_pow>(furthest_dsqd);
}
template<int t_pow>
@@ -62,9 +62,9 @@ class bounds_aux {
furthest_point_in_bound1[d] = bound1_range.hi;
v = v2;
}
furthest_dsqd += std::pow(std::fabs(t_pow), v); // v is non-negative
furthest_dsqd += math::PowAbs<t_pow, 1>(v); // v is non-negative
}
furthest_dsqd = std::pow(2.0/t_pow,furthest_dsqd);
furthest_dsqd = math::Pow<2, t_pow>(furthest_dsqd);
}
template<int t_pow, typename TVector>
@@ -90,8 +90,8 @@ class bounds_aux {
furthest_point_in_bound1.CopyValues(bound1.center());
la::AddExpert(bound1.radius(), unit_vector, &furthest_point_in_bound1);
furthest_dsqd = std::pow(2.0/t_pow,
la::RawLMetric<t_pow>(bound2_centroid.length(),
furthest_dsqd = math::Pow<2, t_pow>
(la::RawLMetric<t_pow>(bound2_centroid.length(),
furthest_point_in_bound1.ptr(),
bound2_centroid.ptr()));
}