This repo contains the code and data for the paper "A Comparison of Statistical Relational Learning and Graph Neural Networks for Aggregate Graph Queries".
To run the experiments, create a directory called data and download the zip file from https://linqs-data.soe.ucsc.edu/public/embar-mlj21.zip. Unzip the files in the data directory
Compile the PSL marginal code to generate the jar file. To do this go to psl directory and run mvn clean install -DskipTests
Copy the generated jar psl-cli-2.1.0.jar to the scripts directory cp ./psl-cli/target/psl-cli-2.1.0.jar ../scripts/
To lean the weights of the psl models run the run_all_wl.sh script bash run_all_wl.sh The learned models are written to the output directory
To generate the samples for the datasets run the run_all_marginal.sh script bash run_all_marginal.sh The generated samples are written to the output directory
To compute the accuracy and AGQs for the samples, run the run_evaluation.sh script bash run_evaluation.sh