Epsilon-Advanced Colour Passing

This repository contains the source code of the epsilon-advanced colour passing algorithm that has been presented in the paper "Approximate Lifted Model Construction" by Malte Luttermann, Jan Speller, Marcel Gehrke, Tanya Braun, Ralf Möller, and Mattis Hartwig (IJCAI 2025).

Our implementation uses the Julia programming language.

Computing Infrastructure and Required Software Packages

All experiments were conducted in a virtual machine running Ubuntu 22.04.2. The code was run on a single core with 32GB RAM available.

We used Julia version 1.8.1 together with the following packages:

• BayesNets v3.4.1
• BenchmarkTools v1.4.0
• CSV v0.10.15
• Clustering v0.15.8
• Combinatorics v1.0.2
• DataFrames v1.6.1
• Distributions v0.25.109
• Graphs v1.9.0
• Multisets v0.4.4
• StatsBase v0.33.21

Moreover, we applied openjdk version 11.0.25 to run the (lifted) inference algorithms, which are provided in the .jar file located at instances/ljt-v1.0-jar-with-dependencies.jar.

Reproducing the Results

After the required software has been installed, the experiments can be started as follows.

Instance Generation

First, the input instances need to be generated. To do so, run julia instance_generator.jl and julia mimic_generator.jl in the src/ directory. To be able to run julia mimic_generator.jl, both patients.csv and procedures_icd.csv need to be present in the instances/mimic/ directory (can be downloaded at https://physionet.org/content/mimiciv/3.1/). The input instances are then written to instances/input/ and to instances/mimic/, respectively.

Running the Experiments

After the instances have been generated, the experiments can be started by running julia run_eval.jl and julia run_mimic.jl in the src/ directory. The (lifted) inference algorithms are then directly executed by the Julia script. All results are written into the results/ directory. To create the plots, run julia prepare_plot.jl in the results/ directory to combine the obtained run times into averages and afterwards execute the R script plot.r (also in the results/ directory). The R script will then create a bunch of .tex files containing the plots of the experiments in the results/ directory. To generate the plots as .pdf files instead, set use_tikz = FALSE in line 7 of plot.r before executing the R script plot.r.