>"HMM-Check has given less than 6 results - using monitoring mode only"
This is related to 'maxBayesValues'.
monitoring only if $this->{hmmres} < int($maxBayesValues / 12 + 1)
scoring only if $this->{hmmres} < int($maxBayesValues / 3 + 1)
If we get so less results for a mail, we can't really say it is spam or
ham. The graph of detection exactness is like below, where X is the number
of results and Y the exactness.
We need at least 7 results - more than 60 gives no better detection. These
values could be a bit different for each user/installation.
_____
/
/
/
/
/
____/
The resulting penalty is calculated in a more or less complex math.
Involved are:
spamprobability:
hamprobability:
combined probability:
bayesian confidence:
corpus confidence:
At the end the valence value is multiplied with the combined probability
to get the penalty.
For example: if the PenaltyMessageLimit is set to 50 and you want a single
(HMM or Bayes) combined probability of 0.9 to block - 56 would be fine as
valence value.
If both checks are enabled, it is on you to calculate the valence values
to your needs. (H, H+B, B)
And to get it a bit more complex: if the rebuild shows: Jan-27-14 04:18:30
Corpus norm: 1.0000 - (very good - balanced)
all is fine. If this value is < 0.9 or >1.1 the result will be lower
penalty values in every case.
my settings are:
PenaltyMessageLimit:=50
baysValencePB:=70
HMMValencePB:=70
forget the locals, I don't scan local mails. But you see, I want to block
anything that smells somehow like spam. I had no false postive (HMM/Bayes)
.. let me see... in the last ~69.500 mails.
I manualy maintain my corpus for less than one hour in a week. And I'm
using several automatic spam collecting (honeypots) addresses with just in
time reporting. So, I get the newest spam very quick, but not on my
account :-)
>eg. divide them by 2. or set the bayes values to 1/3 and the HMM values
to 2/3
I guess this - means you should think about (calculate or try and error) -
you want to know the golden way - go to the Yukon River, I think there are
some nuggets left :-):-):-)
Thomas
Von: Grayhat <[email protected]>
An: [email protected],
Datum: 31.01.2014 14:47
Betreff: [Assp-test] Bayes, HMM and valence
By default the bays and HMM valence values are set to 49 and 55
(regular and local); now... according to the interface if one enables
both checks (setting them to score), the values should be revised...
the problem is that it's unclear how to set them; I mean, if I read the
GUI it says
"for this reason it is recommended to use both Bayesian and HMM. If you
enable both checks, check your settings for baysValencePB,
HMMValencePB, bayslocalValencePB and HMMlocalValencePB - eg. divide
them by 2. or set the bayes values to 1/3 and the HMM values to 2/3."
now... following the above one would set values to (say) 24 and 27 (or
25 and 28) but... is this correct ? Also, I tried enabling HMM (that
is setting it to "score") and then running a rebuild to ensure its table
was correctly populated but even then, ASSP logged messages like this
"HMM-Check has given less than 6 results - using monitoring mode only"
and while I can't understand the above (ok, probably I'm just dumb) the
above means that HMM won't score the message so I'd be left running on
bayes and since I changed bayes to 24, some spam could get through...
Bottom line; what's the correct way (step by step) to enable HMM, how
to correctly set the valence(s) and... what does that "less than..."
message mean ?
------------------------------------------------------------------------------
WatchGuard Dimension instantly turns raw network data into actionable
security intelligence. It gives you real-time visual feedback on key
security issues and trends. Skip the complicated setup - simply import
a virtual appliance and go from zero to informed in seconds.
http://pubads.g.doubleclick.net/gampad/clk?id=123612991&iu=/4140/ostg.clktrk
_______________________________________________
Assp-test mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/assp-test
DISCLAIMER:
*******************************************************
This email and any files transmitted with it may be confidential, legally
privileged and protected in law and are intended solely for the use of the
individual to whom it is addressed.
This email was multiple times scanned for viruses. There should be no
known virus in this email!
*******************************************************
------------------------------------------------------------------------------
WatchGuard Dimension instantly turns raw network data into actionable
security intelligence. It gives you real-time visual feedback on key
security issues and trends. Skip the complicated setup - simply import
a virtual appliance and go from zero to informed in seconds.
http://pubads.g.doubleclick.net/gampad/clk?id=123612991&iu=/4140/ostg.clktrk
_______________________________________________
Assp-test mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/assp-test