MeTeoR: Practical Reasoning in Datalog with Metric Temporal Operators
Authors: Dingmin Wang, Pan Hu, Przemysław Andrzej Wałęga, Bernardo Cuenca Grau5906-5913
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We have implemented this approach in a reasoner called Me Teo R and evaluated its performance using a temporal extension of the Lehigh University Benchmark and a benchmark based on real-world meteorological data. Our experiments show that Me Teo R is a scalable system which enables reasoning over complex temporal rules and datasets involving tens of millions of temporal facts. |
| Researcher Affiliation | Academia | Dingmin Wang1, Pan Hu1,2, Przemysław Andrzej Wał ega1, Bernardo Cuenca Grau1 1Department of Computer Science, University of Oxford, UK 2School of Electronic Information and Electrical Engineering, Shanghai Jiao Tong University, China {dingmin.wang, przemyslaw.walega, bernardo.cuenca.grau}@cs.ox.ac.uk, pan.hu@sjtu.edu.cn |
| Pseudocode | Yes | Algorithm 1: Apply Rules; Algorithm 2: Practical fact entailment |
| Open Source Code | Yes | We have implemented our approach in a new reasoner called Me Teo R (https://meteor.cs.ox.ac.uk) |
| Open Datasets | Yes | LUBM Benchmark. We obtain the first benchmark by extending the Lehigh University Benchmark (LUBM) (Guo, Pan, and Heflin 2005) with temporal rules and data. Meteorological Benchmark. For this benchmark, we used a freely available dataset with meteorological observations (Maurer et al. 2002) |
| Dataset Splits | No | The paper does not explicitly mention the use of a validation set or specific validation splits. It discusses training data implicitly through the Datalog programs and datasets, and testing through input query facts. |
| Hardware Specification | Yes | All experiments have been conducted on a Dell Power Edge R730 server with 512 GB RAM and two Intel Xeon E5-2640 2.6 GHz processors running Fedora 33, kernel version 5.8.17. |
| Software Dependencies | Yes | Our implementation uses existing libraries in the Python 3.8 eco-system without depending on other third-party libraries. Finally, we used Postgresql 10.18 for all our baseline experiments. |
| Experiment Setup | Yes | We have designed two new benchmarks. LUBM Benchmark. We obtain the first benchmark by extending the Lehigh University Benchmark (LUBM) with temporal rules and data. To construct temporal datasets we modified LUBM s data generator... We used the same approach to generate input query facts for entailment checks. We identified five disjoint types of input query facts for an input program Π and dataset D. Meteorological Benchmark. For this benchmark, we used a freely available dataset with meteorological observations (Maurer et al. 2002); in particular, we used a set D푊of 397 millions of facts from the years 1949 2010. We then adapted the Datalog MTL program used by Brandt et al. (2018) to reason with Datalog MTL about weather, which resulted in a a non-recursive program Π푊with 4 rules. |