| =head1 NAME |
| |
| perceptron - Generate scores for SpamAssassin using the "Stochastic |
| Gradient Method" |
| |
| =head1 SYNOPSIS |
| |
| perceptron [options] |
| |
| Options: |
| -p ham_preference Modifies tendency to prefer false negatives over |
| false positives (default 2.0) (higher = less fp) |
| -e num_epochs Set number of passes to make (default 15) |
| -l learning_rate Modifies learning rate (default 2.0) |
| -t threshold Minimum threshold for spam (default 5.0) |
| -w weight_decay Scores multiplied by this value after each pass |
| to prevent scores from getting too high |
| (default off (1.0)) |
| |
| =head1 DESCRIPTION |
| |
| B<perceptron> is used to optimize the SpamAssassin scores. The output |
| of B<logs-to-c> is compiled into the perceptron which is then run to |
| generate scores. |
| |
| =head1 SEE ALSO |
| |
| L<logs-to-c> |
| |
| =cut |