CHANGELOG

v1.2.1

• added binomial logistic regression
• updated to Breeze 0.6

v1.2.0

• breeze-learn pulled into Nak
• K-means from breeze-learn and Nak merged.
• Added locality sensitive hashing

v1.1.3

• Enabled the hashing trick to be used for linear models. See nak.data.HashedExampleIndex and nak.data.HashedFeatureMap.
• PCA now supported directly in Nak for dimensionality reduction w/ kmeans.
• All OpenNLP Maxent code purged from Nak.

v1.1.2

• Incorporated breeze.data classes into nak.data and updated liblinear training and model use to work with these rather than the legacy opennlp classes.
• Added nak.data.Featurizer and nak.core.FeaturizedClassifier, which handle turning raw data into collections of features. When combined with IndexedClassifier, the indexation of these features is handled automatically so that a user of the API doesn't need to ever worry about the low level and can focus on the data and feature extraction.
• Added NakContext object (inspired by SparkContext), which provides a number of utility methods for getting classifiers up and running.
• Added nak.example package, with example implementations for prepositional phrase attachment (PpaExample) and text classification (TwentyNewsGroupsExample).
• Added nak.util.ConfusionMatrix class, which provides detailed error output.
• Added ScalaTest and started writing BDD tests.
• Refactored a lot of code to get rid of duplication.
• Added code documentation to many classes and functions.

v1.1.1

• Massive reorganization of the sub-packages.
• Added nak.liblinear (using the Java liblinear package) and added nak.app.Classify that uses liblinear logistic regression solvers as well as GIS.

v1.1.0

• The classification code from the OpenNLP Maxent package, slightly reorganized.
• The k-means clustering code from Scalabha.
• Changed project id from com.jasonbaldridge to org.scalanlp.

v1.0

• The original OpenNLP Maxent code, pulled out of Chalk and then renamed to nak.*