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

minimized separable Gradient() and and EvaluateWithGradient().

Browse Source
This commit is contained in:
ayesdie
2019-04-07 22:05:07 -04:00
committed by Ryan Curtin
parent 05fba00d8f
commit c55ea016e9
1 changed files with 22 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/mlpack/methods/linear_svm/linear_svm_function_impl.hpp
+22 -20
View File
@@ -313,44 +313,45 @@ void LinearSVMFunction<MatType>::Gradient(
// Scores for each class are evaluated.
arma::mat scores;
// Check intercept condition.
if (!fitIntercept)
{
scores = parameters.t() * dataset;
scores = parameters.t() * dataset.cols(firstId, lastId);
}
else
{
scores = parameters.rows(0, dataset.n_rows - 1).t() * dataset
+ arma::repmat(parameters.row(dataset.n_rows).t(), 1,
dataset.n_cols);
scores = parameters.rows(0, dataset.n_rows - 1).t()
* dataset.cols(firstId, lastId)
+ arma::repmat(parameters.row(dataset.n_rows).t(), 1, dataset.n_cols);
}
arma::mat margin = scores - (arma::repmat(arma::ones(numClasses).t()
* (scores % groundTruth), numClasses, 1)) + delta
- (delta * groundTruth);
* (scores % groundTruth.cols(firstId,lastId)), numClasses, 1))
+ delta - (delta * groundTruth.cols(firstId, lastId));
// For each sample, find the total number of classes where
// ( margin > 0 ).
arma::mat mask = margin.for_each([](arma::mat::elem_type& val)
{ val = (val > 0) ? 1: 0; });
arma::mat difference = groundTruth
arma::mat difference = groundTruth.cols(firstId,lastId)
% (-arma::repmat(arma::sum(mask), numClasses, 1)) + mask;
// Check intercept condition
if (!fitIntercept)
{
gradient = dataset.cols(firstId, lastId)
* difference.cols(firstId, lastId).t();
* difference.t();
}
else
{
gradient.set_size(size(parameters));
gradient.submat(0, 0, parameters.n_rows - 2, parameters.n_cols - 1) =
dataset.cols(firstId, lastId)
* difference.cols(firstId, lastId).t();
* difference.t();
gradient.row(parameters.n_rows - 1) =
arma::ones<arma::rowvec>(batchSize)
* difference.cols(firstId, lastId).t();
* difference.t();
}
gradient /= batchSize;
@@ -440,44 +441,45 @@ double LinearSVMFunction<MatType>::EvaluateWithGradient(
// Scores for each class are evaluated.
arma::mat scores;
// Check intercept condition.
if (!fitIntercept)
{
scores = parameters.t() * dataset;
scores = parameters.t() * dataset.cols(firstId, lastId);
}
else
{
scores = parameters.rows(0, dataset.n_rows - 1).t() * dataset
+ arma::repmat(parameters.row(dataset.n_rows).t(), 1,
dataset.n_cols);
scores = parameters.rows(0, dataset.n_rows - 1).t()
* dataset.cols(firstId, lastId)
+ arma::repmat(parameters.row(dataset.n_rows).t(), 1, dataset.n_cols);
}
arma::mat margin = scores - (arma::repmat(arma::ones(numClasses).t()
* (scores % groundTruth), numClasses, 1)) + delta
- (delta * groundTruth);
* (scores % groundTruth.cols(firstId,lastId)), numClasses, 1))
+ delta - (delta * groundTruth.cols(firstId, lastId));
// For each sample, find the total number of classes where
// ( margin > 0 ).
arma::mat mask = margin.for_each([](arma::mat::elem_type& val)
{ val = (val > 0) ? 1: 0; });
arma::mat difference = groundTruth
arma::mat difference = groundTruth.cols(firstId,lastId)
% (-arma::repmat(arma::sum(mask), numClasses, 1)) + mask;
// Check intercept condition
if (!fitIntercept)
{
gradient = dataset.cols(firstId, lastId)
* difference.cols(firstId, lastId).t();
* difference.t();
}
else
{
gradient.set_size(size(parameters));
gradient.submat(0, 0, parameters.n_rows - 2, parameters.n_cols - 1) =
dataset.cols(firstId, lastId)
* difference.cols(firstId, lastId).t();
* difference.t();
gradient.row(parameters.n_rows - 1) =
arma::ones<arma::rowvec>(batchSize)
* difference.cols(firstId, lastId).t();
* difference.t();
}
gradient /= batchSize;