Revision history for Perl extension Algorithm::NaiveBayes.

 - Fixed a runtime error in the Gaussian model - the code died looking
   for a rescale() function, which I had forgotten to import. [Manju Putcha]

 - Added save_state() and restore_state() methods, which will help
   consumers (like AI::Categorizer) avoid a pesky "can't find method"
   error when using a restored model.

0.03  Mon May 17 22:11:00 CDT 2004

 - The double-loop inside the predict() method has been turned
   inside-out, so that the outer loop is now over the new attributes,
   rather than over the different labels in the training set.

 - Some internal changes for making customization easier: moved the
   model-creation code to new internal classes, making it easier to
   write additional classes that handle model creation differently.

 - Moved some of the numerical utility functions to a
   Algorithm::NaiveBayes::Util module.

 - We now use sum() and max() from List::Util rather than implementing
   our own.  The ones in List::Util are considerably faster.

 - Added a copyright and license statement. [Spotted by Alexey Tourbin]

0.02  Fri May 23 13:36:48 CDT 2003

 - The do_purge() method was called during training regardless of
   whether the 'purge' flag was set.  This has been fixed.

0.01  Tue Mar 18 18:45:34 2003

 - original version; created by extracting the NaiveBayes code out of