Query-driven Qualitative Constraint Acquisition

Authors: Mohamed-Bachir Belaid, Nassim Belmecheri , Arnaud Gotlieb, Nadjib Lazaar, Helge Spieker

JAIR 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental This section presents an experimental evaluation of the revised GEQCA-I, which we refer to as GEQCA-II. We compare GEQCA-II with classical GEQCA-I (Belaid et al., 2022) when learning temporal constraints based on Allen s Interval Algebra with |Γ| = 13 (see Figure 1(a)), and learning spatial networks based on RCC8 with |Γ| = 8 (see Figure 1(b)).
Researcher Affiliation Academia Mohamed-Bachir Belaid EMAIL NILU, Climate and environmental research institute, Kjeller, Norway. Department of Computer Science, Oslo Met, Oslo, Norway. Nassim Belmecheri EMAIL Simula Research Laboratory, Oslo, Norway. Arnaud Gotlieb EMAIL Simula Research Laboratory, Oslo, Norway. Nadjib Lazaar EMAIL LIRMM, University of Montpellier, CNRS, Montpellier, France. Helge Spieker EMAIL Simula Research Laboratory, Oslo, Norway.
Pseudocode Yes Algorithm 1: GEQCA-II: Constraint Acquisition via Qualitative Queries. [...] Algorithm 2: Path-Weighted constraint selection heuristic [...] Algorithm 3: Path-Lex constraint selection heuristic
Open Source Code Yes The code and complete description of each instance can be found at the following public repository: https:// github.com/lirmm/Constraint Acquisition/tree/GEQCA.
Open Datasets Yes We use publicly available RCPSP instances3 and consider the problem structure with task durations, resource requirements, and resource capacities as background knowledge K, denoted by K1. [...] https://github.com/MiniZinc/minizinc-benchmarks/tree/master/rcpsp
Dataset Splits Yes We generated 40 instances per algebra and triplet (n, d, l), with 20 instances being SAT and 20 being UNSAT.
Hardware Specification Yes All tests are run on an Intel core i7, 2.8GHz with RAM of 16GB.
Software Dependencies No The algorithms GEQCA-I and GEQCA-II were implemented in Java, and the Choco solver2 was used for the solve procedure at line 11 in Algorithm 1.
Experiment Setup Yes We use each heuristic with three basic settings: (1) GEQCA-I is used as described in (Belaid et al., 2022) (2) K is empty, and (3) the PC function runs until a fixpoint is reached (cutoff = ). [...] In Table 4, we present the oracle effort for 5 scheduling instances using GEQCA-II and GEQCA-I with the Path-Lex heuristic, a cutofftime of 3, 600s, and different background knowledge (K) settings: , K1, and K1 K2.