Lcap is a small library for active learning of the intersection of several regular languages, on top of active automata-learning libraries. The initial version of Lcap is based on:

• Sebastian Junges and Jurriaan Rot, Learning Language Intersections, 2022

Currently, Lcap is built on top of two libraries:

• dfa / lstar
• aalpy

Run

First, download this repository (or unpack the archive). Then install dependencies with:

pip install -r requirements.txt

Note: We do not currently provide an installation package.

The script lcap-experiments.py allows you to rerun the experiments of [1]. The script produces some outputs on the command line, but more importantly, it creates two files corresponding with the two tables in [1]:

• table1.tex and
• table2.tex

These files contain tables in latex that can be compiled into pdf.

To run all benchmarks once, run

python lcap-experiments.py -b mod -b imod -l indep -l wbw -l mbm

This tells the script to run

• the benchmark-sets mod and imod (via the -b option), and
• the learners
  • independent indep,
  • word-by-word wbw,
  • machine-by-machine mbm

As the learning algorithms are randomized, experiments should be run multiple times:

python lcap-experiments.py -n 20 -b mod -b imod -l indep -l wbw -l mbm

You can also change the seed --seed X or select only a subset of learners (by removing some -l options) and benchmarks (by removing some -b options).

The algorithms are implemented in lcap/learners.py. The lcap-experiments.py script exemplifies the usage. Note that we currently require the membership/equivalence oracles to be based on the actual automaton (ground truth).

In the folder results/ you can find reference results for published experiments.