The OpenNLP Maximum Entropy Package / News: Recent posts

This minor release includes some optimizations to improve model loading times.

This release provides support for real-valued features and for multi-threaded evaluation of models.

Hi all,
There was a bug in the training routine in the initial 2.4.0 release. Since its so soon since the release, I've just replaced the 2.4.0 release with a patched version. If you downloaded in 2.4.0 in the last 10 days, you need to re-download it. Sorry about that and thanks to Paul LaForge for bringing the bug to my attention...Tom

This release includes a change to the model data structure. This dramatically improves memory size when the model is being used as well as cpu time required to make predictions. Model formats have remained unchanged so old models can continue to be used with the new package and in general models produced with the new package should be identical to those produced with the previous version. Special thanks to Richard Northedge for suggesting the data structure changes. Enjoy

Performance optimizations added in this release.

This release consist of a combination of previous work release under the OpenNLP moniker as well as new work. The tools contain a sentence detector, a tokenizer, a pos-tagger, a chunker, a name finder, and a full parser. Each of these components uses the OpenNLP Maxent package.

This release helps with memory requirements when training model with large numbers of events and features. Enjoy.

The latest release contains some important bug fixes to the training algorithm. Models should be retrained with the new version for better performance.

Tom has finally had a chance to have a go at the code after a long while and has fixed up a number of things in the package. It is highly recommended that anyone using previous versions of the package upgrade to the latest version and retrain their models.