Optimised Maintenance of Datalog Materialisations

Authors: Pan Hu, Boris Motik, Ian Horrocks

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We have implemented our hybrid algorithms and have compared them empirically.
Researcher Affiliation Academia Pan Hu, Boris Motik, Ian Horrocks Department of Computer Science, University of Oxford Oxford, United Kingdom firstname.lastname@cs.ox.ac.uk
Pseudocode Yes Algorithm 1 DRED(Π, λ, E, I, E , E+)
Open Source Code Yes Our test system and datasets are available online.1
Open Datasets Yes We used the following benchmarks for our evaluation: UOBM (Ma et al. 2006) ... Reactome (Croft et al. 2013) ... Uniprot (Bateman et al. 2015) ... Chem BL (Gaulton et al. 2011) ... Claros ... and SSPE (Single-Source Path Enumeration).
Dataset Splits No The paper describes testing with small and large deletions but does not specify training, validation, or test dataset splits with percentages or counts.
Hardware Specification Yes We conducted all experiments on a Dell Power Edge R720 server with 256GB RAM and two Intel Xeon E5-2670 2.6GHz processors, running Fedora 24, kernel version 4.8.12-200.fc24.x86 64.
Software Dependencies No The paper mentions the operating system and kernel version ('Fedora 24, kernel version 4.8.12-200.fc24.x86 64') but does not provide specific software dependencies or libraries with version numbers (e.g., Python, PyTorch, TensorFlow, specific solvers).
Experiment Setup Yes All algorithms handle insertions using the semina ıve evaluation. The only overhead is in counter maintenance, which we measured during initial materialisation (which also uses semina ıve evaluation). Hence, the main focus of our tests was on comparing the performance of our algorithms on small and large deletions. In both cases, we first materialised the relevant program on the explicit facts, and then we performed the following tests. To test small deletions, we measured the performance on ten randomly selected subsets E E of 1,000 facts.