This document contains information about the input data and instructions on how to the R2O2* meta-reasoners for OWL as well as how to generate ontology metrics.
The source code expects Java 1.8. It is structured as a Maven project, therefore dependencies should be automatically resolved by Maven.
The following three java files are the entry points to the R2O2* meta-reasoner.
This class is responsible for generating training and testing instances for our meta-reasoning framework R2O2*. The main function in the class is able to generate the following two datasets:
-
ErrorsRemoved: this dataset excludes ontologies that produced runtime (reasoning time) errors by any reasoner: hermit, pellet, more, fact, jfact, and konclude.
-
ErrorsReplaced: this dataset includes the above error ontologies. The reasoning time of these ontologies is the pre-defined timeout threshold, which is 30 mins = 1800000 milli-seconds.
For more information about these datasets, please refer to our paper.
For more instructions on how to generate these datasets, please refer to the above java file. The detailed instructions are given as comments in the java file. Overall, it generates n-fold cross-validation datasets for R2O2*, where n is given as an input. For example, if n = 10, then we generate 0 ~ 9 different directories under the given top directory (i.e. given as an input). For each of these directories, we can see:
-
The
elk/directory: it contains a filer.elk.arffthat will be used to build a prediction model for ELK. -
The
pm/directory: it contains a list of files with names liker.reasoner_name.arff, wherereasoner_nameis the name of a reasoner (e.g., fact, hermit, etc.). In our demonstration, we can see 6 different reasoner names. -
The
test_mm/directory: it contains the following arff files:-
input-pred.arff: It is an arff file for test ontologies, where reasoning time is not given, i.e., all reasoning time is -1 (unknown). This file will be used to evaluate R2O2*pt. The actual reasoning time of different reasoners on these ontologies are found in 'meta-time-actual.arff'. -
input-pred-name.csv: This file contains information about the indices of test ontologies (in order) and their actual names. It shows which ontologies are tested in the current fold. -
meta-rank-actual.arff: This file is used to evaluate R2O2*rk, where each instance shows ontologies' metric values and reasoners' rankings in terms of their efficiency. A lower value indicates a higher rank. -
meta-time-actual.arff: This file is used to evaluate R2O2*rk, where each instance shows ontologies' metric values and actual reasoning time of the reasoners. -
pm-rank.arff: This file is used to evaluate R2O2*mc.
-
-
The
train_mm/directory: It has the same names of the files astest_mmbut the files contain training data.
This class provides functionality for building our 4 different meta-reasoners in the meta-reasoning framework R2O2* (without ELK as one of the component reasoners). For detailed instructions, please see the main() method where detailed comments are provided in the code.
This class provides functionality for evaluating the proposed 4 meta-reasoners with ELK. To evaluate these models, MetaReasonerFramework.java needs to be run first to build key components in each meta-reasoner.
In the given Java package, the data/ directory contains 2 sample datasets (ORE 2015 datasets) used in our paper. The readers can generate training and testing data for cross validation using the code DataGeneration.java. Please see the above explanation under the heading demo/DataGeneration.java.
The entry point to the metrics calculation component is DirectoryProcessor.java in the package edu.monash.infotech.owl2metrics.metrics.jgrapht. It takes a folder (of ontologies) as input and generates a .csv file containing metric values of these ontologies.
Please feel free to contact us if you have questions about the code base and the paper.
- Yong-Bin Kang (ykang@swin.edu.au)
- Yuan-Fang Li (yuanfang.li@monash.edu)