The Quipu Maximum Entropy Package

This is an important release in which a number of corrections were made to the GIS algorithm that improve the speed and accuracy of the training algorithm. Lots of bugfixes were also made.

There is now a smoothing option for training models, but it is not well tested.
It will either improve or degrade your model's performance, depending on how
much data you have and how many classes the model must consider. There is also
a sample available in the source download, named maxent/samples/sports. It
should be very helpful for newbies, but it is also useful as a quick test app
when working on the maxent package itself. A few bug fixes and convenience
classes were added.

LGPL licensing, a new BasicEnglishAffixes class to perform basic morphological stemming for English, a fix for a bug in which the model would not train properly in situations where the number of features was constant, changing the top level package name from quipu to opennlp, lots of little tweaks to reduce memory consumption, a new opennlp.maxent.io subpackage, model training no longer automatically persists the new model and relies on EventStreams rather than EventCollectors, a thread-safe GISModel, and other changes.

The GIS algorithm and the GISModel were reworked to use efficient data structures, giving better memory and speed performance on classification tasks which have many outcomes (without damaging performance for binary decisions). More code documentation was added. This version is fully compatible with models built using v0.2.0.

Initial release.