Consequence-Based Reasoning for Description Logics with Disjunctions and Number Restrictions
Authors: Andrew Bate, Boris Motik, Bernardo Cuenca Grau, David Tena Cucala, František Simančík, Ian Horrocks
JAIR 2018 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In Section 6 we present the results of an extensive performance evaluation, which revealed Sequoia to be competitive with existing reasoners. |
| Researcher Affiliation | Academia | Andrew Bate EMAIL Boris Motik EMAIL Bernardo Cuenca Grau EMAIL David Tena Cucala EMAIL Frantiˇsek Simanˇc ık EMAIL Ian Horrocks EMAIL Department of Computer Science University of Oxford Wolfson Building, Parks Road Oxford, OX1 3QD, United Kingdom |
| Pseudocode | Yes | The pseudocode of our saturation algorithm is shown in Algorithms 2 and 3. |
| Open Source Code | Yes | We have implemented our calculus in a new reasoner called Sequoia.2 ... 2. http://www.cs.ox.ac.uk/isg/tools/Sequoia/ ... 3. http://github.com/andrewdbate/Sequoia/ |
| Open Datasets | Yes | We used the Oxford Ontology Repository5 as our source of test data. ... 5. http://www.cs.ox.ac.uk/isg/ontologies/ |
| Dataset Splits | No | The paper focuses on classifying ontologies and evaluating the performance of reasoners on these ontologies, rather than machine learning tasks that typically involve training/test/validation splits for a model. Therefore, specific dataset split information is not provided as it's not applicable in the context of this research. |
| Hardware Specification | Yes | We run our experiments on a Dell server with 512 GB of RAM and two Intel CPU E5-2640 V3 2.60 GHz processors with eight cores per processor and two threads per core. The system was running Fedora release 26, kernel version 4.11.9-300.fc26.x86 64, and Java 1.8.0 update 151. |
| Software Dependencies | Yes | We used Hermi T 1.3.8, Pellet 2.4.0, Fa CT++ 1.6.5, and Konclude 0.6.2 for tests over the entire corpus. Also, to verify that our algorithm indeed exhibits a pay-as-you-go behaviour, we also compared our system with ELK 0.4.0, Snorocket 2.8.1, and jcel 0.24.1 on the ELH ontologies of our test corpus. ... The system was running Fedora release 26, kernel version 4.11.9-300.fc26.x86 64, and Java 1.8.0 update 151. |
| Experiment Setup | Yes | We allocated 100 GB of heap memory to each Java reasoner, and a maximum private working set of 100 GB for the native code. In order to prevent individual tests from interfering, for each ontology we started a fresh reasoner process that loaded the ontology using the OWL API and then classified it four times. Each classification task was allowed ten minutes to complete, and we measured its wall-clock time; however, since the Java virtual machine was restarted for each ontology, we discarded the first classification task as warm-up. Thus, if all four classification tasks succeeded, we report the average of the last three runs; otherwise, we report the entire test as failed. In all tests we excluded the ontology parsing time in order to analyse the performance of the core reasoning problem. ... Also, we configured Sequoia to use the cautious strategy. |