Skip to content

@mtf90 mtf90 released this Oct 12, 2020

Full changelog

Added

  • A StateLocalInputSULSymbolQueryOracle to wrap a StateLocalInputSULin a SymbolQueryOracle (e.g. to learn partial systems with the ADT learner).
  • Added an example for parallel setups (ParallelismExample1).

Changed

  • The SULSymbolQueryOracle now better handles the pre and post cycles of a SUL (e.g. calls to the reset method now allow to close the embedded SUL from outside).
  • Reworked parallel oracles
    • Several SULOracle variants are no longer thread-safe. This reduces overhead for scenarios where no parallelism is required.
    • The ParallelOracleBuilders factory now offers builder methods for SULs, ObservableSULs, StateLocalInputSULs, MembershipOracless and OmegaMembershipOracles to allow an easy (and correct) construction of parallel setups given one of the mentioned implementations.
  • Refactored the following packages/classes:
    • de.learnlib.oracle.parallelism.ParallelOracleInterruptedException -> de.learnlib.api.oracle.parallelism.BatchInterruptedException
  • The initialPrefixes and initialSuffixes methods of AbstractExtensibleAutomatonLStar are now final since these values can be provided via the constructor of the class. This allows one to simplify sub-classes.
  • Updated to AutomataLib 0.10.0

Removed

  • Removed the learnlib.queries.parallel.threshold property. Learning setups that want to use parallelism now need to explicitly setup parallel oracles.
  • Removed MQUtil#answerQueries{Auto,Parallel} and MQUtil#answerOmegaQueries{Auto,Parallel})
  • LassoOracle#isOmegaCounterExample(boolean) has been removed. This decision can be directly integrated into the #findCounterExample method which has more information available.

Fixed

  • Fixed a bug where NL* would create non-canonical hypotheses (#70 , thanks to @Jaxan)
Assets 2

@mtf90 mtf90 released this Feb 6, 2020

Full changelog

Changed

  • The {DFA,Mealy,}W{p,}MethodEQOracle(MembershipOracle, int, int) constructor no longer interprets its second int parameter as the batch size, but as an estimate for the expected SUL size. In order to explicitly set the batch size of the oracle, use the {DFA,Mealy,}W{p,}MethodEQOracle(MembershipOracle, int, int, int) constructor. Now, the two parameters lookahead and expectedSize will determine the length of the middle part via Math.max(lookahead, expectedSize - hypothesis.size()). This allows to dynamically adjust the length of the middle part throughout the learning process. See LearnLib/automatalib#32.
  • Several DFA/Mealy specific (oracle) subclasses are now automatically generated. As a result they are no longer an inner class but an independent top-level class. This requires to update the import statements.
  • JSR305 annotations have been replaced with checker-framework annotations.
    • LearnLib (incl. AutomataLib) now follows checker-framework's convention that (non-annotated) types are usually considered non-null unless explicitly annotated with @Nullable.
    • LearnLib (incl. AutomataLib) no longer has a (runtime-) dependency on JSR305 (and other javax.*) annotations or includes them in the distribution artifact. This now makes LearnLib (incl. AutomataLib) compliant with Oracle's binary code license and allows LearnLib (incl. AutomataLib) artifacts as-is to be bundled in binary distributions with Oracle's JDKs/JREs.
  • A lot of code for inferring partial Mealy machines (esp. PartialLStarMealy and PartialObservationTable) has been removed/refactored. The concept of state local inputs is now implemented as a SUL filter and introduces a special StateLocalInputSULOracle which early-answers queries that would traverse unavailable inputs with a previously specified symbol. This way, queries that would traverse undefined input symbols still won't be executed on the SUL but the SUL appears as a 'total' Mealy system to the learner, allowing one to use every currently existing Mealy learner as-is. See the in-tree examples for more information.
  • SULCache no longer implements MembershipOracle.
  • Updated to AutomataLib 0.9.0

Removed

  • As a remainder of its initial implementation, the TTTEventListener (and the corresponding event feature in the TTT algorithm) has been removed due to the lack of usage.
Assets 2

@mtf90 mtf90 released this Feb 18, 2019

Full changelog

Added

  • Added support for black-box-checking (thanks to Jeroen Meijer).
  • Added support for learning partial Mealy Machines with state-local inputs via L* (thanks to Maren Geske).
  • Added support for resumable caches.
  • DynamicParallelOracles can now be constructed from a collection of independent oracles.
  • Support for Java 11. Note: LearnLib/AutomataLib still targets Java 8, and thus needs classes provided by this environment (specifically: annotations from javax.annotation). If you plan to use LearnLib/AutomataLib in a Java 11+ environment, make sure to provide these classes. They are not shipped with LearnLib/AutomataLib.

Changed

  • Refactored the following packages/classes:
    • de.learnlib.api.algorithm.feature.SupportsGrowingAlphabet -> net.automatalib.SupportsGrowingAlphabet
    • de.learnlib.api.algorithm.feature.ResumableLearner -> de.learnlib.api.Resumable
  • Some runtime properties for dynamically configuring LearnLib have been renamed. There now exists the LearnLibProperty enum as a single reference point for all available properties.
  • The node iterators for discrimination trees are now hidden behind the DiscriminationTreeIterators factory.
  • Parallel Oracles:
    • The withDefault* methods have been removed from the {Dynamic,Static}PrallelOracleBuilders. If needed, use the regular with* methods and supply the public default values from {Dynamic,Static}PrallelOracle.
    • The new*ParallelOracle methods from the ParallelOracleBuilders factory no longer interpret a single membership oracle parameter as a supplier to a shared oracle, but rather as a single oracle (and thus return a builder for a parallel oracle with fixed pool size).
  • Adding new symbols to learning algorithms (via the SupportsGrowingAlphabet interface) now requires the learner to be initialized with a GrowingAlphabet instance. This is to make sure that the user has full control over which alphabet instance should be used instead of LearnLib making decisions on behalf of the user.
  • Discrimination-Tree based Learners (DT, KV, TTT) now batch queries whenever possible, thus allowing to fully utilize parallel oracles.
  • Also, see the changes in AutomataLib

Fixed

  • Several bugs detected by our ongoing efforts to write tests.
Assets 2

@mtf90 mtf90 released this May 12, 2018

Full changelog

Fixed

  • Fixed an out-of-bounds error in a cache implementation
  • Fixed visibility issues with exportable classes used for the ResumableLearner interface
  • Fixed an issue when adding a new symbol to a learner and the initial alphabet was already an instance of GrowableAlphabet
  • General consolidations (typos, wrong documentation, etc.)
Assets 2

@mtf90 mtf90 released this Feb 27, 2018

Full changelog

Added

  • Added randomized version of W(p)-Method based equivalence oracles (see #40)
  • Added the ADT (adaptive distinguishing tree) active learning algorithm
  • Added two active learning algorithms for visibly pushdown languages.
  • Added the RPNI (regular positive-negative inference) passive learning algorithm, including EDSM (evidence-driven state merging) and MDL (minimum description length) variants.
  • Many active learning algorithms now support adding additional alphabet symbols after initial instantiation/starting of the learning process.
  • Added support for suspending the learning process to a savable / serializable state. The learning process may be resumed from this state at a later point in time.
  • Added the AbstractTestWordEQOracle class, which allows to implement custom equivalence oracles solely based on lazy (stream-based) test-word generation. Existing equivalence oracles (as far as possible) have been reworked to extend this class and thus profit from its built-in laziness and batch (parallelization) support.

Changed

  • Refactored the Maven artifact and Java package structure. Have a look at the List of LearnLib Artifacts for an updated overview of available artifacts. In general, no functionality should have been removed (except of code marked with @Deprecated). The easiest way to migrate your code to the new version is probably by using the Auto-Import feature of your IDE of choice.

    The non-trivial refactorings include:

    • API methods no longer use wildcards in generic return parameters. This allows your code to not having to deal with them.
    • Changes to AutomataLib.
  • Replaced System.out and JUL logging, with calls to a SLF4j facade.

  • Code improvements due to employment of several static code-analysis plugins (findbugs, checkstyle, PMD, etc.) as well as setting up continuous integration at Travis CI.

Fixed

  • Several bugs detected either by our newly employed static code-analysis toolchain or by our ongoing efforts to write tests.
Assets 2