[Spambayes] question about bayes classifier application

whille whille at 163.com
Fri Jan 25 04:34:44 CET 2013


Hi all,
      i've just signed in the email group. please tell me if any mistakes.
      i'm asking suggestion for my testing result.

      i'm using spambayes src codes, to apply in person classifying: 
personal or organization.
      basically use Tester and Bayes codes:

from spambayes.classifier import Bayes
from spambayes.Options import options
from spambayes.Tester import Test, _Example

tokenizing sample, add 'user:' to username field:
         username: user:...
         description: ...
         vip: True/False

the result i got now is appended,  400~4k samples used to train and predict.
method: use 90% to train, rest 10% to test. average 10 times to get result.

5-5-img3.png:
options["Categorization", "ham_cutoff"] = 0.5
options["Categorization", "spam_cutoff"] = 0.5
false_positive_rate still climbing at 4k samples.

4-8.5-....png:
options["Categorization", "ham_cutoff"] = 0.4
options["Categorization", "spam_cutoff"] = 0.85
unsure_rate still climbing. false_positive_rate climbing slowly. If add 
both, result in similar as 0.5-.0.5 cutoff

Is this normal in spambayes app? Any wrong  in my usage? thanks a lot.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: 5-5-img3.png
Type: image/png
Size: 36776 bytes
Desc: not available
URL: <http://mail.python.org/pipermail/spambayes/attachments/20130125/76166896/attachment-0002.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 4-8.5-img-vtype-imgcomplex.png
Type: image/png
Size: 36984 bytes
Desc: not available
URL: <http://mail.python.org/pipermail/spambayes/attachments/20130125/76166896/attachment-0003.png>


More information about the SpamBayes mailing list