Re: BlameThorstenAndJenny: doesn't appear to classify some mails correctly after learning

On 11/6/2009 11:06 AM, Bill Yerazunis wrote:

> The OSB code is also essentially unchanged for the last four years
> (we have changed only things needed to make it 64-bit-safe).  The
> algorithms are unchanged, and the differences in how it works
> _on the test sets_ is within a small fraction of a percent (thanks,
> Kurt!).

I do believe I've had problems for at least that long.  unfortunately, I find 
the interface to be sufficiently messy that it is really difficult to  change 
engines such as winnoning.  I still haven't moved to mailreaver because after 
every call, I would need to delete files from the cache because I do not need an 
unmanaged cache duplicating everything I have a database (and, I found much to 
my surprise that sqlite management of messages are much much faster than 
messages held in the file system especially when your message count crosses 6000 
or 7000).    Multiple writers is a bitch but reading is pretty damn fast.

> Of course, YMMV.  But try a threshold of 5.0 and see how that
> performs for you.

I'll doublecheck it but the last time I tried this it didn't change much.  a 
whole bunch of red messages just piled up under 5.0.  Messages that should have 
been clearly dumped, worse yet, messages that are marked as spam were not being 
recognized.

I really need to spend some time moving the messages into new database and 
re-scoring them by hand and give you a test case showing exactly what's 
happening.  I've had enough people look at my training code that I'm confident 
it's highly likely to be to right  but not confident enough to be a totally 
arrogant coder about it.

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with
Crystal Reports now.  http://p.sf.net/sfu/bobj-july