Revision history for Perl extension Algorithm::SVMLight.

0.09  Tue Nov 25 10:05:46 CST 2008

 - Updated our patch file to account for changes in the latest
   released version of svmlight.

 - Updated our README to fix a couple mismatches in the
   instructions. [Spotted by Andreas Koenig]

 - new() will now die if you pass a parameter it doesn't know about.
   This helps save your skin if you pass "costratio" instead of
   "svm_costratio", for example, and wonder why it's not having any
   effect.

0.08  Mon Jan  7 15:53:51 CST 2008

 - Changed the INSTALL file into a README file with better
   installation instructions.

 - Added the ability to use a customizable Perl cost function when
   learning a ranking SVM.

 - Fixed a comment in param_set.c.PL that was causing some warnings.

 - Added an explicit typecast in SVMLight.xs to silence some compiler
   warnings.

 - Changed VERSION to SVMLIGHT_VERSION in the patch file, to avoid
   conflicts with perl's VERSION macro in the XS file.

0.07  Thu Nov 30 13:58:24 CST 2006

 - Forgot to include the param_set.c.PL file, which meant people
   couldn't build the module.  Looks like this was broken since
   version 0.04.

0.06  Thu Nov 30 13:42:45 CST 2006

 - Added a 'query_id' parameter to add_instance() for ranking SVMs.

0.05  Mon Jul 17 15:25:50 CDT 2006

 - The mapping between feature names and feature indices (the internal
   integers that svmlight uses to represent features) is now preserved
   in model files written & read by write_model() and read_model().

 - The various accessor methods for the internal SVMLight properties
   (get_epsilon_crit(), set_poly_degree(), and so on) are now
   automatically generated by a little perl script.  This fixes some
   properties I didn't have accessors for.

0.04  Mon Apr 24 10:41:07 CDT 2006

 - All the kernel parameters and learning parameters now have get_*
   getters as well as set_* setters.

 - The kernel parameters (kernel_type, poly_degree, ...) can now be
   set by passing parameters to the new(...) method, just like the
   learning parameters.

 - I'm now automatically generating the C code for the getters &
   setters, so they should be less prone to typos or omissions.

0.03  Wed Jan  4 17:02:35 CST 2006

 - Added the get_linear_weights() method, which returns an array of
   the trained weights used in a linear model.

0.01  Thu Jul  7 11:45:21 2005
 - original version