learnlib-19.0.0

Full changelog

Added

• GitHub releases now provide learnlib-cli artifacts for using LearnLib via the command-line interface without the need for Java or Maven.
• Added a new (L*-based) learning algorithm for Mealy machines with local timers (MMLTs), including support for parallel queries, caching, and conformance testing (thanks to @pdev55).
• Added the L^s active learning algorithm for Mealy machines (thanks to @stateMachinist).
• Added an EarlyExitEQOracle which for a given AdaptiveMembershipOracle and TestWordGenerator stops the evaluation of (potentially long) Mealy-based equivalence tests as soon as a mismatch with the hypothesis is detected, potentially improving the symbol performance of the given equivalence oracle.
• Both lambda learners (LLambda{DFA,Mealy} and TTTLambda{DFA,Mealy}) now support the Resumable interface.
• The LLambda{DFA,Mealy} learners now implement the OTLearner interface to export their internal knowledge via an ObservationTable.
• Most regular learner no implement AccessSequenceTransformer as a means to extract representatives of hypothesis states.

Changed

• LearnLib shifted version numbers to more prominently communicate breaking changes with major versions.
• LearnLib now requires Java 17 at runtime.
• Statistics collection has received a major rework. Previously, classes would implement the StatisticCollector interface and return a StatisticData object which 1) only allows for describing a very limited amount of data, and 2) requires you to keep track of all the objects that collect data. This approach has been replaced by a new StatisticsService. Instances of this service can be obtained similar to a logger via Statistics.getService() and require you to provide an implementation of this service on the classpath (a default one is provided by the learnlib-statistics module). The new service allows arbitrary components to collect various data which can be conveniently extracted based on the new StatisticsKeys used by the components. For more details on advanced scenarios (such as multi-threaded benchmarking), see the documentation of the respective classes. While this may require you to adjust the way you are collecting statistics, all functionality from beforehand should still be available.
  • SimpleProfiler has been replaced by the new clock-based statistics.
• Most learners now more rigorously implement the LearningAlgorithm contract that, e.g., duplicate invocations of startLearning or calling refineHypothesis / getHypothesisModel before startLearning throw IllegalStateExceptions.
• Experiment now has type variables for the input symbol type and output domain type.
• {DFA,Mealy,Moore}Experiment have been moved to the de.learnlib.util package.
• The generateTestWords method of AbstractTestWordEQOracle now needs to be public.
• The classes of de.learnlib.testsupport.it.learner have been split into the packages de.learnlib.testsupport.it{,testcase,util,variant} in the same module (de.learnlib.testsupport:learnlib-learner-it-support).
• Switched to AutomataLib 13.0.0.

Removed

• All adapters from the learnlib-procedural learner have been removed due to main learners implementing AccessSequenceTransformer now. Use the constructors of the main learners instead.

Fixed

• The de.learnlib.algorithm.adt module now correctly exports the de.learnlib.algorithm.adt.config.model.* packages.
• The TTTLambdaMealy learner now returns stable hypotheses that no longer issue queries during traversal.

New Contributors

• @pdev55 made their first contribution in #153
• @stateMachinist made their first contribution in #152