dkpro-tc-0.7.0

DKPro TC 0.7.0-Release

Highlights

• Issue 183: support for CRFsuite machine learning framework
• Issue 190: Support for SVMhmm machine learning framework
• Issue 126: new generic machine learning module dkpro-tc-ml, new functionality to set a machine learning adapter
• Issues 199, 200: New functionality to set the feature store, new sparse feature store

Enhancements

• Additional demos for new machine learning frameworks and other experiment setups
• New module dkpro-tc-evaluation for cross-framework evaluation functionality (not fully supported yet, only works for single-label learning mode)

Major Version Enhancements

• Upgrade to DKPro Core 1.7.0 / DKPro parent pom 10
• Upgrade to Weka 3.7.11
• Upgrade to Meka 1.7.3

Other Issues Fixed

• Issue 220: Fixed wrong calculation of label-wise scores in multi-label learning mode
• Issue 163: SimpleDkproTCReader does not allow to set language
• Issue 139: Experiments should fail with a meaningful exception if no data are found
• Issue 160: Invalid column index in reports which produce XLS files
• Issue 175: ValidityCheckTask only verifies the experiment setup based on training data

Other

• Module de.tudarmstadt.ukp.dkpro.tc.features.pair.similarity-asl has been excluded from the Maven Central release, since the "DKPro Similarity" dependency is not yet available on Central.
• All demos in dkpro-tc-examples and dkpro-tc-examples-groovy now have unit tests.
• dkpro-tc-ml-mallet is deprecated and will be removed from the next release due to a lack of support.