dbacl - digramic Bayesian classifier Git

75 lines (73 with data), 1.9 kB

\" t
.TH HYPEX 1 "Bayesian Classification Tools" "Version @VERSION@" ""
.SH NAME
hypex \- computes the Chernoff exponent between two simple categories.
.SH SYNOPSIS
.HP
.B hypex
[-hH size] [-s stepsize]
CATDUMP1 CATDUMP2
.HP
.B hmine
-V
.SH DESCRIPTION
.PP
.B hypex
reads two category dumps produced by dbacl(1) after learning.
A category dump is obtained using the 
.B -d
and
.B -l
switches, and is a textual representation of the feature weights which
exist in the binary category files. 
.PP
Given two such category dumps for simple unigram categories,
.B hypex
calculates the Kullback Leibler divergence between the probability models,
and prints out exponential error exponents for Neyman-Pearson hypothesis 
tests under a range of threshold values. See
.BR "Cover and Thomas (1991) Elements of Information Theory" ,
Chap. 12.
.PP
Note that only simple categories are supported, and the output only
makes sense under appropriate theoretical conditions. This is a very 
specialized calculator.
.SH EXIT STATUS
.B hypex
returns 0 on success, 1 if an error occurs.
.SH OPTIONS
.IP -h
.IP -H
Same as dbacl(1). Selects the hash sizes in powers of two.
.IP -s
Stepsize for the threshold.
.B hypex
outputs exponents for different values of the threshold, within an 
interval bounded by the Kullback Leibler divergences between the categories.
.IP -V
Print the program version number and exit. 
.SH SOURCE
.PP
The source code for the latest version of this program is available at the
following locations: 
.PP
.na
http://www.lbreyer.com/gpl.html
.br
http://dbacl.sourceforge.net
.ad
.SH BUGS
.PP
Doesn't work with complex categories, and theoretical assumptions are
unrealistic in practice. 
.SH AUTHOR
.PP
Laird A. Breyer <laird@lbreyer.com>
.SH SEE ALSO
.PP
.BR dbacl (1), 
.BR mailcross (1),
.BR mailfoot (1),
.BR mailinspect (1),
.BR mailtoe (1),
.BR regex (7)