NAME

bayes.conf - Sets the parameters for generating 'adaptive classification' (Bayesian) spam detection rules


SYNOPSIS

/opt/pmx/etc/spam.d/bayes.conf


DESCRIPTION

The bayes.conf configuration file sets the parameters for generating 'adaptive classification' (Bayesian) spam detection rules (using the command pmx-spam train; see the pmx-spam manpage for instructions). After training, enable adaptive classification rules via the the adaptive.conf manpage configuration file.

datadir
Specify the directory that contains the test messages used to generate rules via training.

min_word_length
The Bayes back end splits the message into words, and calculates the probability of the message being spam based on the frequency of each word in the training data. This option sets the length that a word must be before it counts towards the message's total score.

Words shorter than min_word_length are ignored.

Values: 0 .. infinity

Default: 3

max_word_length
Words longer than max_word_length are ignored.

Values: 0 .. infinity

Default: 12

max_parsed_words
The number of unique words to extract from the message before calculating the message's total score. Once the back end has extracted max_parsed_words, it stops scanning the message.

Values: 1 .. infinity

Default: 50,000

max_scoring_words
The maximum number of words that actually contribute to the message's total score. To save time, only the top max_scoring_words are used.

Values: 1 .. infinity

Default: 20

min_occurrences
The minimum number of times a word must appear in the training set before it will be added to the Bayesian database. This parameter only takes effect during pmx-spam train.

Values: 1 .. infinity

Default: 5


COPYRIGHT

Copyright (C) 2000-2008 Sophos Group. All rights reserved. Sophos and PureMessage are trademarks of Sophos Plc and Sophos Group.