On 05/09/2012 06:04 AM, Stevan Bajic' wrote:
On 09.05.2012 12:45, Tim Streit wrote:


On 05/09/2012 04:32 AM, Stevan Bajic' wrote:
On 09.05.2012 05:23, Tim Streit wrote:
Stevan,
Hello Tim,

Here was the goods from your email from a few hours ago :

X-DSPAM-Result: Spam
X-DSPAM-Processed: Tue May  8 18:11:59 2012
X-DSPAM-Confidence: 0.4821
X-DSPAM-Improbability: 1 in 94 chance of being ham
X-DSPAM-Probability: 0.9921
X-DSPAM-Signature: 1,4fa9a83f269342116717032

I am using the configuration that you'd help me setup- do you think that
I need to adjust any 'threshold' or whatever?
are you telling me that you get to many mails marked as Spam (aka: false
positive)?
Here is the current output :

dspam:
            TP True Positives:                  1477
            TN True Negatives:                  2494
            FP False Positives:                   24
            FN False Negatives:                 1461
            SC Spam Corpusfed:                    37
            NC Nonspam Corpusfed:                  0
            TL Training Left:                      0
            SHR Spam Hit Rate                 50.27%
            HSR Ham Strike Rate:               0.95%
            PPV Positive predictive value:    98.40%
            OCA Overall Accuracy:             72.78%

TL is gone. I understand that in TOE mode, TL running out will not "magically" change the spam coming into the inbox, right (it applies more so to TUM mode where it switches from TEFT-like to TOE-like behaviour?)? What I was trying to say above was false positives had been pretty low (though I just had like 21 of them last night just after emailing you, your own email included), but also the spam being let through so far is pretty high, aka a lot of junk mail getting let through, including a lot of mail thats already been re-trained. See this paste from an email that just came in last night from a user :

Okay. So this leads me to think that you have either a high 'Feature tb=n' value and/or you have a high ''Preference "statisticalSedation=n"' value or your preference value for 'statisticalSedation' is high.
Feature tb=5
statisticalSedation=5

both found in dspam.conf.  Current stats as of the moment :

dspam:
            TP True Positives:                  1826
            TN True Negatives:                  2550
            FP False Positives:                   26
            FN False Negatives:                 1469
            SC Spam Corpusfed:                    43
            NC Nonspam Corpusfed:                  0
            TL Training Left:                      0
            SHR Spam Hit Rate                 55.42%
            HSR Ham Strike Rate:               1.01%
            PPV Positive predictive value:    98.60%
            OCA Overall Accuracy:             74.54%



read the scores - and here is the website - a total spam site - and it
  says a 1 in 255 chance of being spam.  I have put this one in there 25
times in 3 days
http://www.jovalexperiencetime.com/
Message Options :

Received: from ie5.jovalexperiencetime.com (jovalexperiencetime.com 
[108.178.50.76])
  x-priority: 3
  Content-Type: text/plain;
        charset="us-ascii"
  Content-Transfer-Encoding: 7bit
  X-DSPAM-Result: Innocent
  X-DSPAM-Processed: Tue May  8 22:40:15 2012
  X-DSPAM-Confidence: 0.7172
  X-DSPAM-Improbability: 1 in 255 chance of being spam
  X-DSPAM-Probability: 0.0000
  X-DSPAM-Signature: 1,4fa9e71f269347750013885


and another :
I'd be interested in the answer
I have put the same loan email in the spam box 100 times already.  That is what 
I am refering to
Am still ready for you out here.
Just trying to figure out if its common at 4 days old to be having things happen like that, comments come in like that.

If I look at the statistics from the other mail you sent:

dspam:
                TP True Positives:                  1073
                TN True Negatives:                  2153
                FP False Positives:                    3
                FN False Negatives:                 1275
                SC Spam Corpusfed:                    25
                NC Nonspam Corpusfed:                  0
                TL Training Left:                    344
                SHR Spam Hit Rate                 45.70%
                HSR Ham Strike Rate:               0.14%
                PPV Positive predictive value:    99.72%
                OCA Overall Accuracy:             71.63%


Then I see that you are still in training mode (344 left). So you could
either wait till those 344 are down to zero or you can change the
training buffer to a higher value to work around the high false positive
rate.
In dspam.conf increase the value of "Feature tb=n" to a higher value.
Usually n is 5.

After you are out of training mode (aka TL is zero) then you can use the
preference "statisticalSedation" to water down the false positive rate.
It's actually (mostly) the false negatives that have people frustrated?
But you know what? I think you followed my example of creating a MERGED
group but I see that you use a SHARED group. So this is not the same and
therefore my posted preferences for the MERGED group are not optimal for
a SHARED group. Could you please post the output of the following command?
# dspam_admin list preferences dspam
That command returns back nil. I don't mind using a merged group over a shared group, not sure of the pros or cons? You are right, my current configuration is a shared group.
Can you please repeat with the following command:
dspam_admin aggregate preferences dspam
and
dspam_admin aggregate preferences default
root@XXXX:/etc/dspam# dspam_admin aggregate preferences dspam
trainingMode=TOE
spamAction=deliver
spamSubject=[SPAM]
statisticalSedation=5
enableBNR=on
enableWhitelist=on
signatureLocation=headers
tagSpam=off
tagNonspam=off
showFactors=off
optIn=on
optOut=off
whitelistThreshold=5
makeCorpus=off
storeFragments=off
localStore=
processorBias=on
fallbackDomain=off
trainPristine=off
optOutClamAV=on
ignoreRBLLookups=off
RBLInoculate=on
notifications=off

root@XXXX:/etc/dspam# dspam_admin aggregate preferences default
trainingMode=TOE
spamAction=deliver
spamSubject=[SPAM]
statisticalSedation=5
enableBNR=on
enableWhitelist=on
signatureLocation=headers
tagSpam=off
tagNonspam=off
showFactors=off
optIn=on
optOut=off
whitelistThreshold=5
makeCorpus=off
storeFragments=off
localStore=
processorBias=on
fallbackDomain=off
trainPristine=off
optOutClamAV=on
ignoreRBLLookups=off
RBLInoculate=on
notifications=off

You'd mentioned something in one of these emails about my configuration being tuned for a global group, not a shared group? Is there still something there to address do you think?

Thanks,

Tim


Thanks,
Thanks,


------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and
threat landscape has changed and how IT managers can respond. Discussions
will include endpoint security, mobile security and the latest in malware
threats.http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/


_______________________________________________
Dspam-user mailing list
Dspam-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/dspam-user


------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and
threat landscape has changed and how IT managers can respond. Discussions
will include endpoint security, mobile security and the latest in malware
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/


_______________________________________________
Dspam-user mailing list
Dspam-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/dspam-user
------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Dspam-user mailing list
Dspam-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/dspam-user

Reply via email to