mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-11-28 21:20:18 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Added parallel implementation of compute_gradient method.

Browse Source
This commit is contained in:
Eugene Khvedchenya 2016-03-29 11:09:54 +03:00
parent 1e1dc1429f
commit ee92a36123
1 changed files with 38 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

45
modules/ml/src/lr.cpp
View File

@ -362,6 +362,42 @@ double LogisticRegressionImpl::compute_cost(const Mat& _data, const Mat& _labels
return cost;
}
struct LogisticRegressionImpl_ComputeDradient_Impl : ParallelLoopBody
{
const Mat* data;
const Mat* theta;
const Mat* pcal_a;
Mat* gradient;
double lambda;
LogisticRegressionImpl_ComputeDradient_Impl(const Mat& _data, const Mat &_theta, const Mat& _pcal_a, const double _lambda, Mat & _gradient)
: data(&_data)
, theta(&_theta)
, pcal_a(&_pcal_a)
, gradient(&_gradient)
, lambda(_lambda)
{
}
void operator()(const cv::Range& r) const
{
const Mat& _data = *data;
const Mat &_theta = *theta;
Mat & _gradient = *gradient;
const Mat & _pcal_a = *pcal_a;
const int m = _data.rows;
Mat pcal_ab;
for (int ii = r.start; ii<r.end; ii++)
{
Mat pcal_b = _data(Range::all(), Range(ii,ii+1));
multiply(_pcal_a, pcal_b, pcal_ab, 1);
_gradient.row(ii) = (1.0/m)*sum(pcal_ab)[0] + (lambda/m) * _theta.row(ii);
}
}
};
void LogisticRegressionImpl::compute_gradient(const Mat& _data, const Mat& _labels, const Mat &_theta, const double _lambda, Mat & _gradient )
{
@ -379,13 +415,8 @@ void LogisticRegressionImpl::compute_gradient(const Mat& _data, const Mat& _labe
_gradient.row(0) = ((float)1/m) * sum(pcal_ab)[0];
//cout<<"for each training data entry"<<endl;
for(int ii = 1;ii<_gradient.rows;ii++)
{
pcal_b = _data(Range::all(), Range(ii,ii+1));
multiply(pcal_a, pcal_b, pcal_ab, 1);
_gradient.row(ii) = (1.0/m)*sum(pcal_ab)[0] + (_lambda/m) * _theta.row(ii);
}
LogisticRegressionImpl_ComputeDradient_Impl invoker(_data, _theta, pcal_a, _lambda, _gradient);
cv::parallel_for_(cv::Range(1, _gradient.rows), invoker);
}