OrgLion-ML
/
dlib
mirror of https://github.com/davisking/dlib.git
1
0
Fork 0
Code Issues Releases Wiki Activity

Changed the code to normalize the regularizer by the sum of edge weights 

instead of the straight sum of edges.

--HG--
extra : convert_revision : svn%3Afdd8eb12-d10e-0410-9acb-85c331704f74/trunk%403660
This commit is contained in:
Davis King 2010-05-30 19:44:03 +00:00
parent b46186fc21
commit c543a025ba
1 changed files with 16 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
dlib/manifold_regularization/linear_manifold_regularizer.h
View File

@ -26,6 +26,7 @@ namespace dlib
)
{
_size = 0;
sum_edge_weights = 0;
}
struct neighbor
@ -132,10 +133,12 @@ namespace dlib
mutable_blocks[i+1] = mutable_blocks[i] + num_neighbors[i];
}
sum_edge_weights = 0;
// finally, put the edges into data
for (unsigned long i = 0; i < edges.size(); ++i)
{
const float weight = weight_funct(edges[i]);
sum_edge_weights += weight;
// make sure requires clause is not broken
DLIB_ASSERT(weight >= 0,
@ -150,6 +153,12 @@ namespace dlib
}
double sum_of_edge_weights (
) const
{
return sum_edge_weights;
}
private:
/*!
@ -157,10 +166,12 @@ namespace dlib
- _size == 0
- data.size() == 0
- blocks.size() == 0
- sum_edge_weights == 0
CONVENTION
- size() == _size
- blocks.size() == _size + 1
- sum_of_edge_weights() == sum_edge_weights
- blocks == a vector of iterators that point into data.
For all valid i:
- The iterator range [blocks[i], blocks[i+1]) contains all the edges
@ -170,6 +181,8 @@ namespace dlib
std::vector<neighbor> data;
std::vector<const_iterator> blocks;
unsigned long _size;
double sum_edge_weights;
};
}
@ -216,7 +229,7 @@ namespace dlib
impl::undirected_adjacency_list graph;
graph.build(edges, weight_funct);
num_edges = edges.size();
sum_edge_weights = graph.sum_of_edge_weights();
make_mr_matrix(samples, graph);
}
@ -243,7 +256,7 @@ namespace dlib
// sums is typical of most machine learning algorithms. Moreover, doing this makes
// the argument to this function more invariant to the size of the edge set. So it
// should make it easier for the user.
intrinsic_regularization_strength /= num_edges;
intrinsic_regularization_strength /= sum_edge_weights;
return inv_lower_triangular(chol(identity_matrix<scalar_type>(reg_mat.nr()) + intrinsic_regularization_strength*reg_mat));
}
@ -304,7 +317,7 @@ namespace dlib
}
general_matrix reg_mat;
unsigned long num_edges;
double sum_edge_weights;
};
}