Goal and Observation Selection Coverage Experiments presented in the thesis "Explaining the Outcomes of Goal Recognition Systems"

Coverage Experiment for Goal and Observation Selection Source code for the coverage experiments presented in the thesis "Explaining the Outcomes of Goal Recognition Systems." Overview The experiments evaluate the cumulative coverage achieved by: the top-ranked goal selection procedure when selecting the top one, two, and three goals; and the high-impact observation selection procedure when selecting the top one, two, three, and four unique observations. Dependencies The experiments require: R 4.5 or later. The following R packages: arules Ckmeans.1d.dp cluster conflicted fs ggplot2 latex2exp optparse RColorBrewer stringr tidyverse Install all required packages by running the following command in an R console: install.packages(c( "arules", "Ckmeans.1d.dp", "cluster", "conflicted", "fs", "ggplot2", "latex2exp", "optparse", "RColorBrewer", "stringr", "tidyverse" )) Top-Ranked Goal Selection Components The goal selection procedure consists of two scripts: executeDirichletWindowSizeConfidenceLevelExperiments.bat executes dirichlet_windowSize_confidenceLevel_study.R concurrently for each of the five planning domains: blocks-world, logistics, rovers, sokoban, and zeno-travel. Goal probability distributions for all benchmark problems are read from the input directory, and the resulting output files are written to the output directory. dirichletGoalSelectionExperiment.R reads the results stored in the output directory and computes summary coverage statistics. The generated statistics are also written to the output directory. Running the Procedure Open a Windows terminal. Run executeDirichletWindowSizeConfidenceLevelExperiments.bat. Wait for the batch script to finish (this may take several minutes). It generates one RDS file per planning domain in the output directory. Run the main() function in dirichletGoalSelectionExperiment.R. The main() function returns an R list containing the coverage statistics and exports them as a CSV file in the output directory. High-Impact Observation Selection The high-impact observation selection procedure must be executed after the goal selection procedure, as it requires the selected goals as input. Components The procedure consists of a single script: observationImpact.R selects high-impact observations and computes their coverage statistics. It reads the goal selection results from the output directory and the goal probability distributions from the input directory. Running the Procedure Run the mainAllDomains() function in observationImpact.R. The function returns an R list containing the coverage statistics and exports: a CSV file containing the coverage statistics; and a PDF file containing the corresponding plots. Both files are written to the output directory. Citation If you use this code in your research, please cite the associated PhD thesis: @phdthesis{Ferreira2026_ExplainingOutcomesGoalRecognition, title={{Explaining the Outcomes of Goal Recognition Systems}}, author={Ferreira, Jair}, year={2026}, school={{Monash University}} }