Maximum entropy is a powerful method for constructing statistical models of classification tasks, such as part-of-speech tagging in Natural Language Processing. The Quipu Maximum Entropy Package is a Java implementation of the maximum entropy framework. It allows you to train, evaluate, and use maxent models.
|Tags||Scientific/Engineering Artificial Intelligence Mathematics Text Processing Linguistic|
Release Notes: 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.
Release Notes: 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.
Release Notes: 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.
Release Notes: 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.
Release Notes: Initial release.