tommy/libpostal
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[doc] documentation fix for averaged perceptron

Browse Source
This commit is contained in:
Al
2015-09-08 16:37:23 -07:00
parent c80d8b8067
commit 607a607b71
1 changed files with 6 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
src/averaged_perceptron.h
View File

@@ -9,16 +9,17 @@ The averaged perceptron is a linear model, meaning the score for a given class
is the dot product of weights and the feature values. is the dot product of weights and the feature values.
This implementation of the averaged perceptron uses a trie data structure to This implementation of the averaged perceptron uses a trie data structure to
store the mapping from features to ids, which can be quite memory efficient store the mapping from features to indices, which can be quite memory efficient
as opposed to a hash table and allows us to store as opposed to a hash table and allows us to store millions of features with
very little memory.
The weights are stored as a sparse matrix in compressed sparse row format The weights are stored as a sparse matrix in compressed sparse row format
(see sparse_matrix.h) (see sparse_matrix.h)
See [Collins, 2002] Discriminative Training Methods for Hidden Markov Models: Paper: [Collins, 2002] Discriminative Training Methods for Hidden Markov Models:
Theory and Experiments with Perceptron Algorithms Theory and Experiments with Perceptron Algorithms
Paper: http://www.cs.columbia.edu/~mcollins/papers/tagperc.pdf Link: http://www.cs.columbia.edu/~mcollins/papers/tagperc.pdf
*/ */
#ifndef AVERAGED_PERCEPTRON_H #ifndef AVERAGED_PERCEPTRON_H
#define AVERAGED_PERCEPTRON_H #define AVERAGED_PERCEPTRON_H