Improving Delete Relaxation Heuristics Through Explicitly Represented Conjunctions
Authors: E. Keyder, J. Hoffmann, P. Haslum
JAIR 2014 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate the resulting heuristic functions empirically on a set of IPC benchmarks, and show that they are sometimes much more informative than standard delete-relaxation heuristics. ... We evaluate the resulting heuristics on a wide range of benchmarks from the International Planning Competition, varying the relevant algorithm parameters to determine their individual effect on performance. |
| Researcher Affiliation | Academia | Jörg Hoffmann EMAIL Saarland University 66123 Saarbrücken, Germany Patrik Haslum EMAIL The Australian National University & NICTA Canberra ACT 0200, Australia |
| Pseudocode | Yes | Algorithm 1: Choosing C for relaxed plan heuristics. ... Algorithm 2: Choosing C for landmark generation. |
| Open Source Code | No | The paper states: "The compilation and associated heuristics were implemented in the Fast Downward planner (Helmert, 2006)", indicating the use of an existing tool rather than providing code for their specific methodology. No explicit statement of code release or repository link is found. |
| Open Datasets | Yes | We evaluate the resulting heuristic functions empirically on a set of IPC benchmarks, and show that they are sometimes much more informative than standard delete-relaxation heuristics. ... The planners were tested on all of the STRIPS domains from the 1998 2011 editions of the International Planning Competition (IPC). |
| Dataset Splits | No | The paper mentions: "The planners were tested on all of the STRIPS domains from the 1998 2011 editions of the International Planning Competition (IPC). For domains from the last two IPCs, only the most recent sets of instances were used." This indicates the use of pre-defined benchmark instances for evaluation, but does not specify a distinct training/validation/test split created by the authors for their experiments. |
| Hardware Specification | Yes | All experiments were run on Opteron 2384 processors with the settings used in the competition: a memory limit of 2Gb and a time limit of 30 minutes. |
| Software Dependencies | No | The paper states: "The compilation and associated heuristics were implemented in the Fast Downward planner (Helmert, 2006)", but does not specify a version number for Fast Downward itself or any other software libraries. |
| Experiment Setup | Yes | The compilation and associated heuristics were implemented in the Fast Downward planner (Helmert, 2006), and used in a greedy best-first search, with lazy evaluation and a second open list (with boosting) for states resulting from preferred operators. ... All experiments were run on Opteron 2384 processors with the settings used in the competition: a memory limit of 2Gb and a time limit of 30 minutes. ... We tested each value of x from the set {1.5, 2, 2.5, 3} for both ΠC and ΠC ce. ... Conflict selection was based on hmax supporters rather than hadd supporters... |