http://bugzilla.spamassassin.org/show_bug.cgi?id=2910
------- Additional Comments From [EMAIL PROTECTED] 2004-01-21 23:58 ------- Created an attachment (id=1717) --> (http://bugzilla.spamassassin.org/attachment.cgi?id=1717&action=view) 10fcv score files (tar.gz) OK, here's the results from the ten-fold cross-validation run. It's a *little* off, but nothing that isn't tweakable I wouldn't say... looks like a minor issue of balancing FPs vs FNs. Basically the FPs are low at the expense of a few % points of FNs. Attachment is a tar.gz containing the scores files for the GA and perceptron parts of the test. I'll upload a tar.gz with the ham.log and spam.log files as well in a minute. If someone wants, they could take it, tweak the perceptron code to rebalance that FP/FN ratio a bit, and rerun the perceptron half... I documented the procedure at http://wiki.spamassassin.org/w/TenFoldCrossValidation . ga/scores.1 False positives: 36 0.09% (0.19% nonspam) ga/scores.1 False negatives: 846 2.22% (4.42% spam) ga/scores.2 False positives: 40 0.11% (0.21% nonspam) ga/scores.2 False negatives: 827 2.17% (4.32% spam) ga/scores.3 False positives: 44 0.12% (0.23% nonspam) ga/scores.3 False negatives: 856 2.25% (4.47% spam) ga/scores.4 False positives: 39 0.10% (0.21% nonspam) ga/scores.4 False negatives: 865 2.27% (4.52% spam) ga/scores.5 False positives: 41 0.11% (0.22% nonspam) ga/scores.5 False negatives: 813 2.14% (4.25% spam) ga/scores.6 False positives: 40 0.11% (0.21% nonspam) ga/scores.6 False negatives: 869 2.28% (4.54% spam) ga/scores.7 False positives: 36 0.09% (0.19% nonspam) ga/scores.7 False negatives: 825 2.17% (4.31% spam) ga/scores.8 False positives: 44 0.12% (0.23% nonspam) ga/scores.8 False negatives: 817 2.15% (4.27% spam) ga/scores.9 False positives: 39 0.10% (0.21% nonspam) ga/scores.9 False negatives: 873 2.29% (4.56% spam) ga/scores.10 False positives: 44 0.12% (0.23% nonspam) ga/scores.10 False negatives: 862 2.27% (4.51% spam) perceptron/scores.1 False positives: 9 0.02% (0.05% nonspam) perceptron/scores.1 False negatives: 1142 3.00% (5.97% spam) perceptron/scores.2 False positives: 17 0.04% (0.09% nonspam) perceptron/scores.2 False negatives: 1044 2.74% (5.46% spam) perceptron/scores.3 False positives: 22 0.06% (0.12% nonspam) perceptron/scores.3 False negatives: 1067 2.80% (5.58% spam) perceptron/scores.4 False positives: 22 0.06% (0.12% nonspam) perceptron/scores.4 False negatives: 1137 2.99% (5.94% spam) perceptron/scores.5 False positives: 11 0.03% (0.06% nonspam) perceptron/scores.5 False negatives: 1136 2.99% (5.94% spam) perceptron/scores.6 False positives: 21 0.06% (0.11% nonspam) perceptron/scores.6 False negatives: 991 2.60% (5.18% spam) perceptron/scores.7 False positives: 16 0.04% (0.08% nonspam) perceptron/scores.7 False negatives: 1102 2.90% (5.76% spam) perceptron/scores.8 False positives: 22 0.06% (0.12% nonspam) perceptron/scores.8 False negatives: 980 2.58% (5.12% spam) perceptron/scores.9 False positives: 24 0.06% (0.13% nonspam) perceptron/scores.9 False negatives: 972 2.55% (5.08% spam) perceptron/scores.10 False positives: 22 0.06% (0.12% nonspam) perceptron/scores.10 False negatives: 975 2.56% (5.10% spam) ------- You are receiving this mail because: ------- You are the assignee for the bug, or are watching the assignee.
