forked from larsga/Duke
-
Notifications
You must be signed in to change notification settings - Fork 0
ReleaseNotes
jeff-phil edited this page Mar 4, 2014
·
2 revisions
Contains the release notes for older Duke versions from before the move to Github.
See 1.2 Release Notes.
The main new feature in this version is the GeneticAlgorithm, which can be used to automatically tune a configuration.
Other new features:
- Support for geosearch.
API changes:
- Moved Column class to .datasources.
- Property is now an interface, and no longer a class.
- Added setDoInference to !InMemoryLinkDatabase.
- Added !ConfigWriter, which can export configurations to XML.
- Added !LinkFileWriter.
- Added Processor.setProfiling and Processor.getProfiler to allow performance profiling via API.
- Added Processor.removeMatchListener
- Added Configuration.copy() and Property.copy().
Changes to command-line tools:
- Added the RecordSearch tool.
- Added --reindex option to DebugCompare.
- Added --lookups to .Duke
Bug fixes:
- SPARQL data source was broken. Now fixed, and test cases added.
- Issue 124: --interactive fails in Eclipse
- Issue 114: --noreindex causes matches to disappear
This is a major new release of Duke. After 18 months of work it is now feature-complete enough to deserve a 1.0 release.
Performance improvements:
- Support for multi-threading added
- Using NIOFSDirectory on all platforms except Windows
- New in-memory backend, faster than Lucene (experimental)
Changes to [Comparator]s:
- Geo-coordinate comparator added.
- Q-grams comparator added.
- Levenshtein implementation is now faster
- Weighted Levenshtein weight estimator now knows position in string (issue 81)
Changes to [Cleaner]s:
- Added PhoneNumberCleaner
- Extended and generalized regexp cleaner
- Removed sub-cleaner concept, added support for multiple cleaners
Other improvements:
- Implemented user control over lookup props
- Upgraded to Lucene 4.0
- Added MatchListener.startProcessing() callback
- Removed some MatchListener callback methods (weren't thread-safe)
- InMemoryLinkDatabase now complete and tested
- LinkDatabaseMatchListener bug fixes
- Better validation of configurations
- JDBCEquivalenceClassDatabase added
- RDBMSLinkDatabase performance improvement
Changes to command-line client:
- Added data debug mode
- Fixed bug with reusing link file as test file
- Added pretty-printing of records
- Better interactive debugging behaviour
- Improvements to DebugCompare tool
- Added performance profiling to command-line client
Bugs fixed:
- Issue 83: Look up record by ID when ID is a URI.
- Issue 90: Bug in command-line option parser.
- Bug in CSV data source fixed
The following changes have been made in this version:
- A change to the calculation of property probabilities when values do not match exactly. This means that you may need to adjust the probabilities and thresholds in your applications.
- Upgraded to Lucene 3.6.1.
- Improvements to NorwegianCompanyNameCleaner and NorwegianAddressCleaner.
This version adds the following features:
- A weighted Levenshtein comparator.
- A Metaphone comparator.
- A Jaccard index comparator.
- A prototype of a comparator using a Norwegian version of Metaphone.
- A generic value cleaner.
- Support for setting objects as parameters of other objects.
This version has seen the internals of the code refactored and cleaned up. In addition, the following has been added:
- Support for pluggable backends in addition to the Lucene backend
- A naïve in-memory backend that matches all pairs of records
- Performance tuning options in the configuration
- Data sources can now split a single value into several properties
- New cleaners: RegexpCleaner, MappingFileCleaner
- The NTriples data source now supports incremental loading
- Properties can now be declared as type="ignore"
- Command-line tool has an --interactive option
- Command-line tool has an --noreindex option
- Levenshtein comparator has been substantially optimized
- Several bug fixes
- Many more tests
- In-memory link database implementation
- inferLink() method on link database interface
This version of Duke introduces:
- Added JNDI data source for connecting to databases via JNDI (thanks to FMitzlaff).
- In-memory data source added (thanks to FMitzlaff).
- Record linkage mode now more flexible: can implement different strategies for choosing optimal links (with FMitzlaff).
- Record linkage API refactored slightly to be more flexible (with FMitzlaff).
- Added utilities for building equivalence classes from Duke output.
- Made the XML config loader more robust.
- Added a special cleaner for English person names.
- Fixed bug in NumericComparator (issue 66)
- Uses own Lucene query parser to avoid issues with search strings.
- Upgraded to Lucene 3.5.0.
- Added many more tests.
- Many small bug fixes to core, NTriples reader, ec.
This version of Duke introduces:
- Refactored API to make it much more user-friendly.
- Documented API to same end.
- New comparators: !NumericComparator, !DiceCoefficientComparator, !SoundexComparator
- A new record linkage mode which can be used to link records from different data sets.
- Numerous bug fixes and new test cases.
- Performance improvements in the Levenshtein comparator.
- Default cleaner now strips accents.
- Upgraded to Lucene 3.3.0.
- Version stamping in manifest file, API, and command-line client.
This version of Duke fixes a number of bugs, and adds a number of improvements.
- Example data and setup now included in distribution.
- New comparators: !JaroWinklerTokenized, !DifferentComparator
- A new !DebugCompare command (see TuningGuide)
- More flexibility in the CSV data source
- Better reporting of configuration errors
- A --verbose option