Published March 14, 2023 | Version v1
Software Open

Benchmarks, Code and Data from Mugdan et al, ICAPS 2023

  • 1. University of Basel

Description

This bundle contains all benchmarks, code, and data used in the paper
"Optimality Certificates for Classical Planning" published at ICAPS 2023.

Benchmarks
Contains two folders:
 - original: The original benchmark set, which is a clone of
   https://github.com/aibasel/downward-benchmarks. Note that we do not use the
   full benchmark set but only a subset used for optimal planning.
 - unsolvability-compilation: The compilation to unsolvability we created with
   the script "cost_parser.py" (see under "Code").

Code
Contains the following folders and scripts:
 - cost-extractor.py: A script to extract optimal cost for tasks from a
   properties file created by a Downward Lab experiment.
 - cost_parser.py: A script that takes the information generated from
   cost-extractor.py and performs the compilation to unsolvability as described
   in the paper. (Only supports unit-cost.)
 - downward-optimality-certificates: A modified version of Fast Downward
   (https://github.com/aibasel/downward/) that we used to create optimality
   certificates.
 - downward-unsolvability: A clone of
   https://github.com/salome-eriksson/downward-unsolvability used for generating
   unsolvability certificates.
 - helve-optimality: A modified version of Helve (see below) that implements
   rules for deriving lower bounds.
 - helve-unsolvability: A clone of https://github.com/salome-eriksson/helve/
   used for verifying unsolvability certificates.

Data
Contains three folders:
 - optimality / unsolvability: A copy of all scripts and logs of the
   experiments run with Downward Lab (https://lab.readthedocs.io/en/stable/).
   Experiments consist of:
   - A script detailing how the experiment is to be conducted (benchmark set,
     code, limits, ...)
   - custom parsers
   - data/<experiment-name>: Contains the logs of each run
   - data/<experiment-name>-eval: Contains "properties", a JSON file that
     summarizes the experiment based on the parsed attributes.
     (For the lmcut experiment only the properties file is present.)
 - analysis: Contains html reports and .tex plots, as well as the scripts that
   generated them.

Files

mugdan-et-al-icaps2023-benchmarks.zip

Files (498.3 MB)

Name Size Download all
md5:c057fe2226b0ca45c533b10d8a247be4
45.9 MB Preview Download
md5:cbd289cfe54b058a23c338a3a4094734
135.8 MB Preview Download
md5:78ee456196424f3f47b39965037a01dd
316.6 MB Preview Download

Additional details

Funding

Certified Correctness and Guaranteed Performance for Domain-Independent Planning (CCGP-Plan) 200021_182107
Swiss National Science Foundation
TAILOR – Foundations of Trustworthy AI - Integrating Reasoning, Learning and Optimization 952215
European Commission