Spami Traing/Testing

[dalek1999]

I have recently installed Spami 9.9.52 to try it out. I finally figured out the basic Training and Recycle scheme - IMHO not very intuitive.

Question:
Is there any way to feed the Training with bulk lots of known Spam and Non Spam?

I have loads of folders in my email clients with thousands of example Spams and Non-Spam messages. Some Spam killer programs let you drag and drop or copy messages into their Training modules - it certainly makes training a lot faster and more accurate.

thx
John

[Chactory]

Hi John,

thank you for your contribution!

Could you explain to us, why you don't take the training window for intuitive?

No need to train Spamihilator with spam! Just place the Learning Filter (Bayesian) at the end of the filter sequence, and the Spam-Word Filter at the second last position. The Spam-Word Filter will sieve probably the most spam, until the Bayesian Filter is trained sufficiently, which happens mostly after only few days.

Drag&Drop is a very interesting idea, but I'm not aware about the possibly very high amount of programmer work to realize that ...

Sincerely, Chactory

[dalek1999]

Hi Chactory

It just does not "feel" as slick as some programs to work with. Because it works as proxy with no extension/addin to email client you need to keep flipping to training area. It seems awful slow to learn and you need to keep marking individual mails.

For testing and training there needs to be some way (import drag n drop etc.) to dump a whole lot of messages into Spami and say these are examples of spam/non spam.

I am not clear why Spami puts copy of messages into Recycle area before user has classified it?

Once you PreMark messages in Training area how can you unmark them all again?

Testing the effectiveness of filters and their sequence is awkward. There is no way I can see to get Spami to re-evaluate filters on messages in Training area so you can see effect of any filter changes you have made.

thx
John

[Chactory]

Hi John!

Thank you for explaining your reasons!

The idea behind the training and recycle windows is, that you don't get all the spam into your mail client, for you can handle the spam detection outside of your regular mail client. And you don't need to watch the Spamihilator windows all the time, as soon as Spamihilator has reached its best recognition performance. I work with the training window, because it pre-marks all the mails within as spam or non-spam. So seldom I have false positive filterings, that I don't need to check the training window often. The recycle window is nearly never needed to look into, apart from the seldom occurring problem, when I had deleted too quickly a false positive recognized mail from the training window.

Actually you don't need to dump a whole lot of messages into Spami to prepare it, because the Spam-Word Filter is prepared to recognize a lot of spam until the Learning Filter will forereach it.

Yes, I too wish that Spamihilator had about three standardized profiles too chose from, and you would not need to work out your own filter sequence and check the success with the statistics plugin. But you have got us, the forum freaks, to help you a little bit. Please try out the hyperlinks in my signature to learn more. I'd like to share my filter sequence with you, so I prepared some screen shots. In the english forum, Zaxon wrote a very helpful posting.

Regards, Chactory

[dalek1999]

Hi Chactory

Thanks for your pointers. I had already checked out your website (very nice) and most of the posts and links on the English forum.

IMHO most users will find it tricky configuring filters as there is lack of clear and detailed info on how they work - also for the reasons I already mentioned it is not possible to do a test on say a set of 500 messages in Training area, change some filter settings then check effectiveness again on same set - it all comes down to suck it and see on new mail!

With regard the Bayes and Word filters common terms are fine but Spam training is still essential. One persons Spam is anothers Ham.

I understand how Spami is designed but I still have few queries on some of its operations:-

What are the exact conditions/timing for mail moving from Training area to client? Sometimes I get msg in client with no copy left in Training area and other times in both. Is behaviour a function of the plugin/filter? I note the Newsgroup filter moves NonSpam straight thru. Other msgs seem to move only once they have been tagged *and* Learnt then pulled by next read request from client.

thx
John

[Chactory]

Hi John,

thanks again for your contribution! And thaks a lot for mentioning my screen-shots with such polite words.

As I answered before, I agree with you when you ask for a completed configuration which prevents user from configuring around a lot. And I think that every user needs to adjust only some very specific details, e.g. I can't use the image filter, for the company where I work sends mail-signatures containing an image loading from a server.

Two filters can learn, the Bayesian Filter and the Link Filter. In general, it needs only three to seven days until those filters have learned enough, and before they can learn, the DCC-Filter and the Spam-Word Filter will catch the biggest part of the spam-mails. So you really don't need to feed Spamihilator with test-mails.

I wonder why you miss "clear and detailed info on how they work", but perhaps I'm already stuck in the routine and don't recognize gaps and traps any more ... A mail received will be checked by every filter in the filter sequence, until one of them decides non-spam or spam. I have configured the Newsletter Filter and the Signature Filter to recognize a mail as non-spam, the Link Filter and the Learning Filter to decide both, spam or non-spam. A mail decided non-spam will be delivered to the mail client and mirrored in the training area, except when you have activated "automatically learn from my friends list" or "automatically learn from my blocked senders list" of the Learning Filter, and "don't save newsletters in the training area".

Sincerely, Chactory