Re: Mailreaver/mailtrainer TOE
Jeff Rice <[email protected]> Sun, 24 Aug 2014 11:39:41 -0400
| Newsgroups | gmane.mail.spam.crm114 |
|---|---|
| Message-ID | <[email protected]> |
This is a multi-part message in MIME format. --===============1867851587768297731== Content-Type: multipart/alternative; boundary="------------040401020309040209090202" This is a multi-part message in MIME format. --------------040401020309040209090202 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit OK, after perusing the "CRM Revealed" book and delving into mailtrainer.crm (and adding a bunch of output statements to follow the flow), I do think mailtrainer.crm implements DSTTTR training. If the message is over the threshold, it is not learned. If it is in the unsure zone (or misclassified), a learn command is executed. If it *still* doesn't score over the threshold, a learn "refute" command is run on the opposite bin. (ie, if it's spam, it is refuted against nonspam) If the classifier is hyperspace, the learn refute action is replaced by a simple repeat of the learn. It would be nice if the header added to the file actually reflected what was going on. Having a header that says "LEARNED AND CACHED GOOD" when in fact that did not happen is misleading. The training report is more accurate, although again I think it should say clearly when a learn command wasn't executed because the message was already over the threshold. Perhaps this was obvious to most people, but it wasn't to me! (and due to the lack of responses, not obvious to at least a few others!) Jeff > Jeff Rice <mailto:[email protected]> > August 22, 2014 at 1:56 PM > > To clarify, I'm not talking about retraining. This would be a "fresh" > message as far as CRM goes - it's never seen it before. But from what > I've read and recall from prior discussions, CRM is most effective > when trained using TOE. So if it already correctly classifies a new > message, I want to make sure I don't try to retrain using it. > > It was simple enough to add this check to my automated script to > prevent this. But things get convoluted and hard to maintain when so > many checks take place in so many different places! > ------------------------------------------------------------------------------ > Slashdot TV. > Video for Nerds. Stuff that matters. > http://tv.slashdot.org/ > _______________________________________________ > Crm114-general mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/crm114-general > Jeff Rice <mailto:[email protected]> > August 22, 2014 at 12:15 PM > Hi, > I realize this list appears to be comatose, but hopefully it's not > completely unresponsive! > > My question is related to mailreaver.crm & mailtrainer.crm, and TOE. > It's hard to tell from the docs and I'm not fluent enough in CRMese to > decipher from the code if TOE training is implemented. Put simply, if I > tell mailreaver to train using a message it classifies as very spammy, > what happens? (mailreaver.crm --spam < spammy.msg). > > When I tested this on message getting a score of -127, the score of the > message didn't change after training and my spam.css feature count was > unchanged. That suggests to me that no training took place. However, > the "documents learned" counter incremented by one, and a header was > added saying "LEARNED AND CACHED AS SPAM". Those imply, at least, that > it was learned as spam but (perhaps?) that no new tokens were found. > (but would have been learned, had any existed) Does this make sense? > > Primarily, I'm thinking about this in the context of a spamtrap on my > server, where all received emails at that address are fed into > Spamassassin and CRM to be learned as spam without user intervention. I > wonder if I need to put a check in before CRM to make sure the message > isn't ALREADY classified as spam before telling CRM to learn it. > > Anyway, I hope people are still using CRM114. I had great results with > it back in the 2007-8 or so but stopped using it for reasons I can't > exactly recall. Recently I added it back in using the Spamassassin > plugin and it's working great, with a few minor hiccups. (lovely > program but a bit temperamental sometimes!) > > Jeff > > > ------------------------------------------------------------------------------ > Slashdot TV. > Video for Nerds. Stuff that matters. > http://tv.slashdot.org/ > _______________________________________________ > Crm114-general mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/crm114-general --------------040401020309040209090202 Content-Type: multipart/related; boundary="------------000804000607000705090108" --------------000804000607000705090108 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit <html><head> <meta content="text/html; charset=ISO-8859-1" http-equiv="Content-Type"> </head><body bgcolor="#FFFFFF" text="#000000">OK, after perusing the "CRM Revealed" book and delving into mailtrainer.crm (and adding a bunch of output statements to follow the flow), I do think mailtrainer.crm implements DSTTTR training.<br> <br> If the message is over the threshold, it is not learned. If it is in the unsure zone (or misclassified), a learn command is executed. If it *still* doesn't score over the threshold, a learn "refute" command is run on the opposite bin. (ie, if it's spam, it is refuted against nonspam) If the classifier is hyperspace, the learn refute action is replaced by a simple repeat of the learn.<br> <br> It would be nice if the header added to the file actually reflected what was going on. Having a header that says "LEARNED AND CACHED GOOD" when in fact that did not happen is misleading. The training report is more accurate, although again I think it should say clearly when a learn command wasn't executed because the message was already over the threshold.<br> <br> Perhaps this was obvious to most people, but it wasn't to me! (and due to the lack of responses, not obvious to at least a few others!)<br> <br> Jeff<br> <br> <blockquote style="border: 0px none;" cite="mid:[email protected]" type="cite"> <div style="margin:30px 25px 10px 25px;" class="__pbConvHr"><div style="display:table;width:100%;border-top:1px solid #EDEEF0;padding-top:5px"> <div style="display:table-cell;vertical-align:middle;padding-right:6px;"><img photoaddress="[email protected]" photoname="Jeff Rice" src="cid:[email protected]" name="compose-unknown-contact.jpg" height="25px" width="25px"></div> <div style="display:table-cell;white-space:nowrap;vertical-align:middle;width:100%"> <a moz-do-not-send="true" href="mailto:[email protected]" style="color:#737F92 !important;padding-right:6px;font-weight:bold;text-decoration:none !important;">Jeff Rice</a></div> <div style="display:table-cell;white-space:nowrap;vertical-align:middle;"> <font color="#9FA2A5"><span style="padding-left:6px">August 22, 2014 at 1:56 PM</span></font></div></div></div> <div style="color:#888888;margin-left:24px;margin-right:24px;" __pbrmquotes="true" class="__pbConvBody"> <meta http-equiv="Content-Type" content="text/html; charset=ISO-8859-1"> <br> To clarify, I'm not talking about retraining. This would be a "fresh" message as far as CRM goes - it's never seen it before. But from what I've read and recall from prior discussions, CRM is most effective when trained using TOE. So if it already correctly classifies a new message, I want to make sure I don't try to retrain using it.<br> <br> It was simple enough to add this check to my automated script to prevent this. But things get convoluted and hard to maintain when so many checks take place in so many different places!<br> <div>------------------------------------------------------------------------------<br>Slashdot TV. <br>Video for Nerds. Stuff that matters.<br><a class="moz-txt-link-freetext" href="http://tv.slashdot.org/">http://tv.slashdot.org/</a></div><div>_______________________________________________<br>Crm114-general mailing list<br><a class="moz-txt-link-abbreviated" href="mailto:[email protected]">[email protected]</a><br><a class="moz-txt-link-freetext" href="https://lists.sourceforge.net/lists/listinfo/crm114-general">https://lists.sourceforge.net/lists/listinfo/crm114-general</a><br></div></div> <div style="margin:30px 25px 10px 25px;" class="__pbConvHr"><div style="display:table;width:100%;border-top:1px solid #EDEEF0;padding-top:5px"> <div style="display:table-cell;vertical-align:middle;padding-right:6px;"><img photoaddress="[email protected]" photoname="Jeff Rice" src="cid:[email protected]" name="compose-unknown-contact.jpg" height="25px" width="25px"></div> <div style="display:table-cell;white-space:nowrap;vertical-align:middle;width:100%"> <a moz-do-not-send="true" href="mailto:[email protected]" style="color:#737F92 !important;padding-right:6px;font-weight:bold;text-decoration:none !important;">Jeff Rice</a></div> <div style="display:table-cell;white-space:nowrap;vertical-align:middle;"> <font color="#9FA2A5"><span style="padding-left:6px">August 22, 2014 at 12:15 PM</span></font></div></div></div> <div style="color:#888888;margin-left:24px;margin-right:24px;" __pbrmquotes="true" class="__pbConvBody"><div>Hi,<br>I realize this list appears to be comatose, but hopefully it's not <br>completely unresponsive!<br><br>My question is related to mailreaver.crm & mailtrainer.crm, and TOE. <br>It's hard to tell from the docs and I'm not fluent enough in CRMese to <br>decipher from the code if TOE training is implemented. Put simply, if I <br>tell mailreaver to train using a message it classifies as very spammy, <br>what happens? (mailreaver.crm --spam < spammy.msg).<br><br>When I tested this on message getting a score of -127, the score of the <br>message didn't change after training and my spam.css feature count was <br>unchanged. That suggests to me that no training took place. However, <br>the "documents learned" counter incremented by one, and a header was <br>added saying "LEARNED AND CACHED AS SPAM". Those imply, at least, that <br>it was learned as spam but (perhaps?) that no new tokens were found. <br>(but would have been learned, had any existed) Does this make sense?<br><br>Primarily, I'm thinking about this in the context of a spamtrap on my <br>server, where all received emails at that address are fed into <br>Spamassassin and CRM to be learned as spam without user intervention. I <br>wonder if I need to put a check in before CRM to make sure the message <br>isn't ALREADY classified as spam before telling CRM to learn it.<br><br>Anyway, I hope people are still using CRM114. I had great results with <br>it back in the 2007-8 or so but stopped using it for reasons I can't <br>exactly recall. Recently I added it back in using the Spamassassin <br>plugin and it's working great, with a few minor hiccups. (lovely <br>program but a bit temperamental sometimes!)<br><br>Jeff<br><br><br>------------------------------------------------------------------------------<br>Slashdot TV. <br>Video for Nerds. Stuff that matters.<br><a class="moz-txt-link-freetext" href="http://tv.slashdot.org/">http://tv.slashdot.org/</a><br>_______________________________________________<br>Crm114-general mailing list<br><a class="moz-txt-link-abbreviated" href="mailto:[email protected]">[email protected]</a><br><a class="moz-txt-link-freetext" href="https://lists.sourceforge.net/lists/listinfo/crm114-general">https://lists.sourceforge.net/lists/listinfo/crm114-general</a><br></div></div> </blockquote> </body></html> --------------000804000607000705090108 Content-Type: image/jpeg; x-apple-mail-type=stationery; name="compose-unknown-contact.jpg" Content-Transfer-Encoding: base64 Content-ID: <[email protected]> Content-Disposition: inline; filename="compose-unknown-contact.jpg" /9j/4AAQSkZJRgABAQEARwBHAAD/2wBDAAEBAQEBAQEBAQEBAQEBAQEBAQEBAQEBAQEBAQEC AQEBAQEBAgICAgICAgICAgICAgICAgICAgICAgICAgICAgL/2wBDAQEBAQEBAQICAgICAgIC AgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgICAgL/wAAR CAAZABkDAREAAhEBAxEB/8QAGAAAAwEBAAAAAAAAAAAAAAAABgcICQr/xAA0EAABAwMCAgUK BwAAAAAAAAACAQMEBQYRABITIQcUMUF2CBUXIjI2N0JRtVRWkZOV0dL/xAAYAQEAAwEAAAAA AAAAAAAAAAADAAEEAv/EACQRAAICAAQGAwAAAAAAAAAAAAABAhEDMrHREyExM0FxgfDx/9oA DAMBAAIRAxEAPwDuEt+gW/ULet6oVC3rfqNQqFv0OfPn1GhUqfOmzZtKZlS5UqZMaNwzNwiJ VIl7eXLCaZIGwBl3TY8epPx2+jy2ZNPjvkwc9uhW8j7nCPhvOsQliYIeS7cvCpp8o50qwrC4 v3lsNSDbdmTEhvs2tahxpfV3WnmbbozJEw/gwdadbYExVRXKEKoSdvJcaOSqxE7/AAiX0gXx +a69/JSf9alIlste0VzaNpeFrcT9KKymotyiaZ0KRCnzacoE7Kjzn4gi2KqUh3jqDHDHv4mR UfruTWlMzlVUKIVNp9GguEJnAh0+IZjyAiisgyRDnu5azS8miKqjOTVkKqS/psG37fo1Fbab eg25b8eZPeFJBBJSjMG5HjMeyihnaauZwe4OGiju13GAcpOwBeN+U8/IkGbsiS8b7ryogmbz hbyc9REROfZhERO5ETShjPtvpGqTUyLErytS4siSwx5x2tRH4hPOI0DkjZtaJtFxuVEbIUUi yeNujlBUJGbJN6nM/Cyf2Hf60YgjvKA+NPSP4gT7axpcPtr51YWJnYn9dnAQWl722p4ot37y zqnlfp6FrqbwawG8/9k= --------------000804000607000705090108-- --------------040401020309040209090202-- --===============1867851587768297731== Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Disposition: inline ------------------------------------------------------------------------------ Slashdot TV. Video for Nerds. Stuff that matters. http://tv.slashdot.org/ --===============1867851587768297731== Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Disposition: inline _______________________________________________ Crm114-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/crm114-general --===============1867851587768297731==--