Re: Efficiency issues/thoughts

[email protected] said:
> Er, yeah, I'm a dunce.  840K.

'course, my guess was highly educated,

-rw-rw-r--    1 jhbrown  user       840390 Aug 27 10:28 /home/ai/jhbrown/.idata

:)

[email protected] said:
> If I were to take a crack at this, would that be productive, or do you
> prefer to do the core coding yourself? 

By all means, go ahead and code away! :)  The chances of me having time 
to do serious work on ifile in the near future are minimal (I'm in the 
middle of proposal writing).

Btw, like you mentioned, it would be nice to pull out blocks of encoded 
information.  Might even be worth doing a minimal MIME parse.  Also, in 
case you didn't see it, take a look at Chris Brown's post from a 8/23.  
Summary points were:

-> On the one hand, performance on a large corpus gets pretty painful.  All the
   word stats get read through, whether they get used or not.
-> It would also be nice to do more sophisticated "symbol tagging."

[email protected] said:
> I've gotten totally addicted to being able to put the first message of
> a thread into its own class, and have the rest of the messages which
> show up over the day turn up in that class.  I can't go back now ;-) 

Understandable.  It's just all too fun to see :)

[email protected] said:
> (What amplifies the problem is that what I think is the "right way" to
> hook ifile into gnus has me invoking ifile twice for every message ---
> once with -q to get a recommendation, and later with -i to assert the
> final decision.) 

Try -Q.  It gives a recommendation and adds the message to the database in
that category.  You can delete (-d) and insert (-i) using a single ifile
(ifile -d spam -i ifile ~/Mail/spam/43) call if that decision is incorrect.

Jason D. M. Rennie
MIT AI Lab
[email protected]
(617) 253-5339
http://www.ai.mit.edu/~jrennie/