AI-DecisionTree

Revision history for Perl extension AI::DecisionTree.

Make a new release so that a META.yml file is generated. [Suggested by Alexandr Ciornii]
as_graphviz(%args) wasn't passing %args through properly to the GraphViz->new() method. Fixed. [Ian Knopke]
get_result() can now take a callback subroutine, rather than just a hash reference, to check the attributes of the given instance. This allows lazy instance checking.

0.08 Mon Jul 7 18:01:16 CDT 2003

0.07 Fri Jun 6 10:37:51 CDT 2003

Created tests for the set_results() and copy_instances() methods.
Added documentation for as_graphviz() and increased the information contained in the GraphViz object.
Added the ability to limit the absolute depth of the tree when training.

0.06 Wed Sep 18 13:59:24 EST 2002

Fixed an XS memory leak that was afflicting all training instances. Added tests to make sure leak stays plugged.
Added the 'purge' and 'verbose' parameters to new().
add_instance() now accepts a 'name' parameter.
Users can now control whether training instances are purged after training, using the 'purge' parameter to new() and/or the do_purge() method.
Added the set_results() and copy_instances() methods, which let you re-use training instances from one tree to another.
Added the instances() and purge() accessor methods.

0.05 Thu Sep 12 01:22:34 AEST 2002

Fixed a concurrency problem that occurred when making more than one decision tree. All tree data is now stored as member data, not class data.
DecisionTree.pm is now pure-perl again (though Instance.pm still has an XS component).
Fixed a one-off bug in the Instance.xs code that could create garbage data.
Handles "sparse" data better. Sparse data means that every attribute doesn't have to be defined for every training/test instance. This can now be a meaningful property - the absence of a value is currently equivalent to a special "<undef>" value.
Don't trigger warnings when undefined attribute values are encountered (as happens with sparse data).
Added documentation for the 'prune' parameter to new()
More consistent with memory allocation in Instance.xs - uses the perl memory macros/functions from `perldoc perlclib` instead of raw malloc/realloc/free.
Catches possible infinite loop situations when growing the tree (which shouldn't usually happen, but other mistakes can cause it)
The Instance class now has a numeric-only interface, without string translation. String translation is done in the main DecisionTree class. This isn't really a user-visible change.

0.04 Wed Sep 4 19:52:23 AEST 2002

Now uses regular XS instead of Inline for the C code parts. [patch by Matt Sergeant]
Converted the inner loop of the best_attr() method to C code, because it was spending a lot of time in accessor methods for the C structures it was using. Don't worry, I'm not going C-crazy. I won't be making many (any?) more of these kinds of changes, but these ones were probably necessary.
Removed a bit of debugging code that I left in for 0.03.

0.03 Mon Sep 2 11:41:18 AEST 2002

Added a 'prune' parameter to new(), which controls whether the tree will be pruned after training. This is usually a good idea, so the default is to prune. Currently we prune using a simple minimum-description-length criterion.
Training instances are now represented using a C struct rather than a Perl hash. This can dramatically reduce memory usage, though it doesn't have much effect on speed. Note that Inline.pm is now required.
The list of instances is now deleted after training, since it's no longer needed.
Small speedup to the train() method, achieved by less copying of data.
If get_result() is called in a list context, it now returns a list containing the assigned result, a "confidence" score (tentative, subject to change), and the tree depth of the leaf this instance ended up at.
Internally, each node in the tree now contains information about how many training examples contributed to training this node, and what the distribution of their classes was.
Added an as_graphviz() method, which will help visualize trees. They're not terribly pretty graphviz objects yet, but they're visual.

0.02 Sat Aug 10, 2002 21:02 AEST

Added support for noisy data, currently by picking the best (most common) result when noise is encountered. See the 'noise_mode' parameter to new().
Added the rule_tree() method, which returns a data structure representing the tree. [James Smith]
Significantly sped up the train() method, especially for large data sets.
The get_result() method is no longer implemented recursively, which simplifies it and speeds it up.
Reformatted the documentation and added a TO DO section.
Documented the nodes() method.

0.01 Sat Jun 8 12:45:03 2002