BUSTLE: a Versatile Tool for the Evolutionary Learning of STL Specifications from Data

This is the official repository for the Evolutionary Computation paper "BUSTLE: a Versatile Tool for the Evolutionary Learning of STL Specifications from Data", hosting all the code necessary to replicate the experiments. This work is mostly based on Federico Pigozzi's master's thesis.

Scope

By running

mkdir output
java -cp libs/JGEA.jar:libs/moonlight.jar:libs/jblas-1.2.4.jar:target/STLRuleEvolutionaryInference.jar it.units.malelab.learningstl.Main {args}

Where {args} is a placeholder for the arguments you must provide (see below), you will launch a grammar-based evolutionary optimization of the formula structure and parameters for Signal Temporal Logic rules. At the same time, several evolution metadata will be saved inside the output folder. The project has been tested with Java 14.0.2.

Warning

On more recent Linux versions, dynamic libraries for jblas may not link properly. In this case, run the following command:

unzip dlibs.zip

to unzip a directory containing the necessary .so files, then

sudo mv dlibs/libgfortran.so.3.0.0 /usr/local/lib/
sudo mv dlibs/libquadmath.so.0.0.0 /usr/local/lib/
sudo ldconfig

To move the .so files to the appropriate place and configure them.

Structure

• src contains all the source code for the project;
• libs contains the .jar files for the dependencies (see below);
• grammars contains the .bnf files with the grammars;
• data contains three datasets;
• target contains the main .jar file;
• dlibs.zip is an emergency kit for the aforementioned warning.

Dependencies

The project relies on:

• JGEA for the evolutionary optimization;
• MoonLight for monitoring signal-temporal logic formulae.

The corresponding jars have already been included in the directory libs. See pom.xml for more details on dependencies.

Usage

This is a table of possible command-line arguments:

Argument	Type	Optional (yes/no)	Default
seed	integer	no	-
dataset	string	no	-
output	string	yes	./output/
mod	{anomaly,supervisde}	no	-
alpha	float	no	-
local	{true,false}	no	-
evolver	{bustle,roge,random}	yes	bustle
threads	integer	yes	# available cores on CPU

The description for each argument is as follows:

• seed: the random seed for the experiment;
• dataset: the name of the dataset;
• output: a (relative) path to the directory to save output files into;
• mod: the modality (semi- or fully-supervised);
• alpha: the alpha penalty;
• local: apply local search or not;
• evolver: the EA for the experiments;
• threads: the number of threads with which to perform evolution. Defaults to the number of available cores on the current CPU. Parallelization is taken care of by JGEA, which implements a distributed fitness assessment.

Bibliography

Please cite as:

@article{pigozzi2024bustle,
  title={BUSTLE: a Versatile Tool for the Evolutionary Learning of STL Specifications from Data},
  author={Pigozzi, Federico and Medvet, Eric and Nenzi, Laura},
  journal={Evolutionary Computation},
  year={2024},
  publisher={MIT Press}
}