MELT 3.1
New
- MELT now has an evaluation command line interface (MELT Evaluation CLI, module
matching-eval-cli
) - three new matchers which uses NLP transformers:
TransformersFilter
,TransformersFineTuner
,TransformersFineTunerHpSearch
- example project which uses NLP transformers
- added negative sampling methods like:
AddNegativesViaMatcher
,AddNegativesRandomlyShare
,AddNegativesRandomlyOneOneAssumption
,AddNegativesRandomlyAbsolute
MatcherHTTPCall
can call remote matchers which implement the web interface defined at the MELT documentationMatcherDockerFile
automatically imports a saved docker image from file into the local repository and starts a fresh container based on this image to call the matcher viaMatcherHTTPCall
- automatic generating of documentation page of all matchers and filters in the MELT repository MELT documentation
- multisource matching with new interfaces and implementations to reuse one to one matchers for a multisource setting
Fixed
- Test Cases of tracks are now ordered in a deterministic way.
getFirstTestCase()
, for instance, will now return
the same test case on every machine. - MatcherSEALS checks before run if the java executable is really Java 1.8 because this is a strict requirement of SEALS client
ConfidenceFinder
in matching-eval returns correct threshold (due to a change of the confidence of false negatives)- MatcherSEALS works also in parallel mode (when executed with
ExecutorParallel
)
Improvements
- Docker Web Packages follow the convention that the file name of the package must carry the name of the image
(with an optional-latest
postfix). The constructor ofMatcherDockerFile
has been adapted so that it is
sufficient to provide a file. - Docker Web packaging plugin: Image name now included in name of package.
- Releasing MELT is now easier because gpg plugin can be activated via a maven profile
- MatcherSEALS uses toString method to return a meaningful name which also includes the name of the exeucted matcher
- The SEALS packaging is now also tested in the CI (tests on all operating systems and java versions)
ConfidenceFinder
also allows to search for best confidence for F-measure, and F beta-measure- Caching of python pip packages, OAEI data, and transformers models to improve CI pipeline runtime
- added transformation route from URL to OWLOntology (OWLAPI) - thus OWLAPI also works with IMatcher interface