Re: size of variables in hyperspace?

[email protected] (Bill Yerazunis) Wed, 10 Jun 2009 09:25:58 -0400 (EDT)
Newsgroups gmane.mail.spam.crm114
Message-ID <20090610132558.F15AE3DE2B6@starbuck>
   From: Thomas Michael Hagen <[email protected]>

   bill, a question about the hyperspace method: does it take the
   ordering of features into account, or could i give it a list of all
   the words in an article in no particular order and get the same
   result?

Order matters in the default configuration.

It matters only because the hyperspace front end uses OSB 
feature generation to create features.  The features are then
sorted, so the old two-finger-merge algorithm can generate 
distances in time O(n+m) rather than O(n*m).  You pay the cost of
the sort only during a LEARN, and the cost of the sort on
the unknowns is only O(n log n) so it's not bad.

Of course, if you put hyperspace into UNIGRAM mode, then yes,
the list of words in no particular order would be equivalent.

    - Bill Yerazunis

------------------------------------------------------------------------------
Crystal Reports - New Free Runtime and 30 Day Trial
Check out the new simplified licensing option that enables unlimited
royalty-free distribution of the report engine for externally facing 
server and web deployment.
http://p.sf.net/sfu/businessobjects