http://bugzilla.spamassassin.org/show_bug.cgi?id=3417





------- Additional Comments From [EMAIL PROTECTED]  2004-05-26 14:54 -------
Than you for patch, Theo!

>2) got 0 hits for me after ~3k mails. 

It strange, because [EMAIL PROTECTED] flag on this rule :)
You did't received messages from us with first 3k emails?

Lets define "Effectivly of Tested Rule" or "Strong of Rule"
(I suppose that this criteria reject or commit new rules)

The first and main criteria is ham/spam ratio for whitelist rules
and spam/ham ratio for blacklist rules.

The second criteria is "popularity" or "wide" - ham/totalhams from whitelist
rules and spam/totalspams for blacklist rules.

For better quality all coefficients must be > 200.

There are rules that have biggest Ratio (big scores) but work seldom.
There are rules that have small Ratio (small scores) but they work almost in
every message and there are many rules of this type.

The total "Rule Strong" I define as production Ratio*Wide

Let see on my rule in your corpus 
Ratio - 92/15 = 6  it very good! but 15<200 (low level of accuracy)
Wide  - 92/172 = 0.5  - very good  92<200

SPF Pass
Ratio  - 15/0 = undefined 15<200, and 0<200 - low level of accuracy
Wide -  15/172 = 0.09,   not bad  15<200, 172<200

Total Strong My Rule in your corpus = 6*0.5 = 3
Total Strong SPF - undefined.

I think SPF very good rule, but not wide now.
Let check our whitelist rules whith Bayes00 and Bayes01!

On my server in russia I think I will have on SPF Pass 0 hams and 0 spams.
zero divide zero = ?

IF SPF became popular, Wide will rise, virus-spamers will send mail correctly
(you are right!), and Ratio will be fall down.

And Total Wide*Ratio will be about constant!

*But if we will be have many whitelist rules (as many as blacklist now()) total
effect from all of them will be very strong.

Whitelist and blacklist rules with the same Strong Wide*Ratio have the equal
possibility to divide mail into spam and ham.

I think SpamAssasin developers should use "Wide*Ratio" as a main criteria to
accept or reject new rules (and remove old rules). 
And here, there is no diffrence between whitelist and blacklist rules.

I remain in opinion, that if we will create "Wide*Ratio" rating for whitelist
and blacklist rules, my rule will be in top20.









------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

Reply via email to