1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
|
\" t
.TH BAYESOL 1 "Bayesian Classification Tools" "Version @VERSION@" ""
.SH NAME
bayesol \- a Bayes solution calculator for use with dbacl.
.SH SYNOPSIS
.HP
.B bayesol
[-DVNniv] -c
.I riskspec
[FILE]...
.HP
.B bayesol
-V
.SH DESCRIPTION
.PP
.B bayesol
is a Bayes solution calculator designed to combine the output of
.BR dbacl (1)
with a prior distribution and a risk specification, and
calculate the optimal Bayesian decision (which minimizes the posterior
risk).
.PP
The risk specification is read from the text file
.I riskspec
and must be written in a simple format described below. The
.BR dbacl (1)
output can either be read from FILE or from STDIN.
.SH EXIT STATUS
On success,
.B bayesol
returns a positive integer corresponding to the category with the lowest risk.
In case of a problem,
.B bayesol
returns zero.
.SH OPTIONS
.IP -c
Classify using
.IR riskspec .
See the section RISK SPECIFICATION.
.IP -i
Fully internationalized mode. Forces the use of wide characters internally,
which is necessary in some locales. This incurs a noticeable performance penalty.
.IP -n
Print risk scores for each
.IR category .
Each score is (approximately) the logarithm of the expected risk under that category. The lowest score (ie closest to -infinity) is best, etc.
.IP -N
Print recursive risk scores for each
.IR category .
Each score is (approximately) the logarithm of the best score based on
the remaining categories, after the previously best scoring categories
have been removed, and a normalizing factor was added. A full
description is given in the technical report listed at the end of this
manpange. The largest score (ie closest to +infinity) is best, etc.
.IP -v
Verbose mode. Prints to STDOUT the category with minimum posterior risk.
In case several categories are possible,
prints the first category in the order in which they appear
in the categories section of
.IR riskpspec .
.IP -D
Print debug output. Do not use.
.IP -V
Print the program version number and exit.
.SH RISK SPECIFICATION
.B bayesol
needs to read a text file
.I riskspec
containing a risk specification. The format of this text file is as follows
.IP
.na
categories {
.IR cat1 ,
.IR cat2 , "" ...,
.IR catN }
.br
prior {
.IR p1 ,
.IR p2 , "" ...,
.IR pN }
.br
loss_matrix {
.br
"\fIregex1\fR" \fIc1\fR [
.IR formula11 ,
.IR formula12 , "" ...,
.IR formula1N ]
.br
"\fIregex2\fR" \fIc2\fR [
.IR formula21 ,
.IR formula22 , "" ...,
.IR formula2N ]
.br
.
.br
.
.br
"\fIregexM\fR" \fIcM\fR [
.IR formulaM1 ,
.IR formulaM2 , "" ...,
.IR formulaMN ]
.br
}
.br
.ad
.PP
In the above,
.IR cat1 ,
.IR cat2 , "" ...,
.IR catN ,
are category names,
.IR p1 ,
.IR p2 , "" ...,
.IR pN ,
are non-negative numbers,
.IR regex1 ,
.IR regex2 , "" ...,
.IR regexM ,
are (possibly empty) regular expression strings,
.IR c1 ,
.IR c2 , "" ...,
.IR cM ,
are instances of the category names
.IR cat1 ,
.IR cat2 , "" ...,
.IR catN ,
and the formulas are numbers or mathematical expressions.
.PP
Every category which appears in the categories section must appear at least
once in the loss_matrix section, with an empty "" regular expression.
To construct the actual loss matrix used in the decision calculations,
.B bayesol
selects, for each category appearing in the categories section,
the first row whose regular expression is matched
within FILE or STDIN, or the first row with empty regular expression if there
are no matches.
.PP
Each formula can be either a single number, or an algebraic combination of
the operators exp(), log(), +, -, *, /, ^ and parentheses (). The string "inf"
is parsed as the value infinity. Also, the
string "complexity" is recognized, and converted to the complexity for
that category
as reported by
.BR dbacl (1).
Finally, if the
corresponding regular expression contains submatches delimited by parentheses,
their numerical values can be used inside the formulas as the special variables
$1, ..., $9. Note that submatches which aren't numerical are converted to the value zero.
.PP
Case is important. Spaces and newlines can be liberally inserted. Comments
must start with a # and extend to the end of the line.
.SH USAGE
.PP
Typically,
.B bayesol
is used together with
.BR dbacl (1).
An invocation looks like this:
.PP
.na
% dbacl -c one -c two -c three sample.txt -vna | bayesol -c toy.risk -v
.ad
.PP
See @PKGDATADIR@/doc/costs.ps for a description of the algorithm used.
See also
@PKGDATADIR@/doc/tutorial.html for a more detailed overview.
.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 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)
|