Re: Some questions setting up latest CRM114

Leonard Lin <[email protected]>
Newsgroups gmane.mail.spam.crm114
Message-ID <[email protected]>
On Apr 4, 2008, at 11:49 AM, Bill Y wrote:

>>   Also, I'm using the 'hyperspace unique' classifier (it seemed  
>> like it
>>   had a lot less unsures early on - is it more accurate overall?)  
>> w/ the
>>   default pR threshold of 10 (I'm using 20070810-BlameTheSegfault).
>>   I've trained 276 Ham and 172 Spam (in date order, ToE and on  
>> UNSUREs)
>>   - as of right now it looks like most new mail is still going in the
>>   unsure folder still - is this normal?  At what point does it  
>> usually
>>   level off?
>
> It should be pretty good now.
>
> ...
>
> I'm finding that a pR threshold of +/- 5.0 is working better these
> days, as it allows micro spams to be caught very easily (microspams
> are spams with just a line of text and single URL as the payload)

Thanks for the tip I'll take a look at the thresholds and see what .   
Do you know offhand about whether the X-CRM114 headers are all ignored  
when processing in mailreaver?  I'm looking at the reaver/lib sources  
and see some mangle calls that I think might be doing the right thing  
but I'm not sure.


Also, and I guess I'll ask it here instead of crm114-discuss since  
there doesn't seem to be a lot of active responses there, but I'm  
interested in also using crm114 for classifying other datasets.

* Has anyone used crm114 to successfully w larger (say, 1-10M item)  
datasets?  Does it blow up either on insert (is it linear speed or  
does it slow down?) or for classifying?

* And secondly, the hyperspace classifier seems really interesting  
because it sounds like (in theory) it would be good for different  
types of classification - I've seen some search results for n-way  
classification -- are there docs/examples for that?

* Also, are there semantics for pulling "relatedness" (i.e. semantics  
for finding distance, or retrieving nearest neighbors)?


.l

-------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.