AP Class Reference

Averaged Perceptron Classifier. More...

#include <ap.h>

Inheritance diagram for AP:

Public Member Functions
void	train (Cases &cases, int T)
	Train the perceptron with a set of `cases`, performing `T` iterations through them.
int	train (Case &cas)
	Train a single case.
Y	predict (X &x)
std::vector< Float >	scores (X &x)
virtual size_t	size ()
Public Attributes
unsigned	k
	k = \|Y\|
unsigned	d
	d = \|X\|
int	t
	t = current number of iterations
std::vector< std::string >	labels
Tanl::Text::WordIndex	predIndex
Static Public Attributes
static float	updatePercent = 0.1F
	Termination threshold.
static bool	verbose = false
	Control output.
static int	partitionSize = 500000
	Split cases into partitions of this maximum size.
Protected Member Functions
virtual void	update (X &x, unsigned yt, std::vector< unsigned > &E)=0
virtual Float	score (unsigned i, X &v)=0
	Similarity score.
	AP (int k=0, int d=0)
virtual bool	load (std::istream &is)
	Load parameters from stream.

Detailed Description

Averaged Perceptron Classifier.

See also:: http://jmlr.csail.mit.edu/papers/volume3/crammer03a/crammer03a.pdf K. Crammer, Y. Singer, Ultraconservative Online Algorithms for Multiclass Problems. Journ. of Machine Learning Research, 2003.

virtual bool AP::load ( std::istream & is ) [protected, virtual]

Load parameters from stream.

Reimplemented in APS, APSV, and APD.

Y AP::predict ( X & x )

References score().

vector< Float > AP::scores ( X & x )

References score().

virtual size_t AP::size ( ) [inline, virtual]

Reimplemented in APS, and APSV.

int AP::train ( Case & cas )

Train a single case.

train(<xi, yi>) train the current model on <xi, yi>

References score(), and t.

The documentation for this class was generated from the following files: