Improving Domain-Independent Planning via Critical Section Macro-Operators
Authors: Lukáš Chrpa, Mauro Vallati7546-7553
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Usefulness of macros is evaluated on several stateof-the-art planners, and a wide range of benchmarks from the learning tracks of the 2008 and 2011 editions of the International Planning Competition. ... Experimental Results The purpose of this experimental analysis is to i) evaluate planners performance on Critical Section macros as well as their combination with MUM (both conservative and aggressive versions), ii) compare them against related state-of-the-art techniques, MUM (Chrpa, Vallati, and Mc Cluskey 2014) and Blo Ma (Chrpa and Siddiqui 2015) and iii) analyse impact of quality of training plans on generated macros utility . We considered domains from the learning track of IPC 2008 and 2011. |
| Researcher Affiliation | Academia | Luk aˇs Chrpa Faculty of Electrical Engineering Czech Technical University in Prague & Faculty of Mathematics and Physics Charles University in Prague Mauro Vallati School of Computing and Engineering University of Huddersfield |
| Pseudocode | Yes | Algorithm 1 Learning Critical Section Macros from training plans |
| Open Source Code | No | The paper does not provide any explicit statement about making the source code for its methodology available, nor does it include any links to a code repository. |
| Open Datasets | Yes | We considered domains from the learning track of IPC 2008 and 2011. |
| Dataset Splits | No | For both methodologies, we considered 6 training tasks per each domain such that their plan length was mostly within 40-80 actions1. Both methodologies consider one training plan per a training task. The MUM methodology uses the same planner for generating training plans as for solving testing tasks. The paper describes training and testing tasks, but does not explicitly mention a separate validation set or split for hyperparameter tuning or early stopping. |
| Hardware Specification | Yes | All the experiments were conducted on Intel Xeon E5 2.0 Ghz, Debian 9. |
| Software Dependencies | No | We have selected 6 state-of-the-art planning engines, according to their results in recent IPCs, and to the exploited planning techniques, namely: LAMA (Richter and Westphal 2010), Probe (Lipovetzky et al. 2014), Mp C (Rintanen 2014), Yahsp3 (Vidal 2014), FDSS 2018 (Seipp and R oger 2018) and Dual BFWS (Lipovetzky et al. 2018). While specific planning engines are mentioned with their publication years (implying versions), the paper does not list specific versions for ancillary software libraries or frameworks like Python, PyTorch, etc., which are typically needed for reproducibility. |
| Experiment Setup | Yes | For each testing task time limit of 900 seconds and memory limit of 4 GB is applied (as in the learning tracks of IPCs). ... For both methodologies, we considered 6 training tasks per each domain such that their plan length was mostly within 40-80 actions. ... The threshold for underrepresented macros was set to 6 (as the number of training tasks). |