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