Awesome

Reasonable

An OWL 2 RL reasoner with reasonable performance

Performance

Comparing performance of reasonable with OWLRL and Allegro. Evaluation consisted of loading Brick models of different sizes into the respective reasoning engine and timing how long it took to produce the materialization. reasonable is about 7x faster than Allegro and 38x faster than OWLRL on this workload.

How to Use

Command Line Interface

You can download a static build of the command line tool here (this is automatically built from the latest master).

Usage as follows:

$ reasonable -h
An OWL 2 RL reasoner with reasonable performance

Usage: reasonable [OPTIONS] <INPUT_FILES>...

Arguments:
  <INPUT_FILES>...  

Options:
  -o, --output-file <OUTPUT_FILE>  [default: output.ttl]
  -h, --help                       Print help
  -V, --version                    Print version

Example:

$ reasonable example_models/ontologies/Brick.n3 example_models/small1.n3 -o myout.ttl
[2023-07-04T15:31:52Z INFO  reasonable] Loading file example_models/ontologies/Brick.n3
[2023-07-04T15:31:52Z INFO  reasonable::reasoner] Loaded 14803 triples from file example_models/ontologies/Brick.n3
[2023-07-04T15:31:52Z INFO  reasonable] Loading file example_models/small1.n3
[2023-07-04T15:31:52Z INFO  reasonable::reasoner] Loaded 14 triples from file example_models/small1.n3
[2023-07-04T15:31:52Z INFO  reasonable] Starting reasoning
[2023-07-04T15:32:11Z INFO  reasonable] Reasoning completed in 18.66sec
[2023-07-04T15:32:11Z INFO  reasonable] Writing to myout.ttl

Python

To facilitate usage, we use the pyo3 project to generate Python 3.x bindings to this project. Installing these should be as easy as pip install reasonable.

See also the brickschema package for working with Brick models. The package provides a generic interface to this reasoner and several others.

Usage looks like:

import reasonable

# import triples from an rdflib Graph
import rdflib
g = rdflib.Graph()
g.parse("example_models/ontologies/Brick.n3", format="n3")
g.parse("example_models/small1.n3", format="n3")

r = reasonable.PyReasoner()
r.from_graph(g)
triples = r.reason()
print("from rdflib:", len(triples))

# import triples from files on disk
r = reasonable.PyReasoner()
r.load_file("example_models/ontologies/Brick.n3")
r.load_file("example_models/small1.n3")
triples = r.reason()
print("from files:", len(triples))

Rust

See Rust docs

Example of usage from Rust:

use ::reasonable::owl::Reasoner;
use std::env;
use std::time::Instant;
use log::info;

fn main() {
    env_logger::init();
    let mut r = Reasoner::new();
    env::args().skip(1).map(|filename| {
        info!("Loading file {}", &filename);
        r.load_file(&filename).unwrap()
    }).count();
    let reasoning_start = Instant::now();
    info!("Starting reasoning");
    r.reason();
    info!("Reasoning completed in {:.02}sec", reasoning_start.elapsed().as_secs_f64());
    r.dump_file("output.ttl").unwrap();
}

OWL 2 Rules

Using rule definitions from here.

TODO: implement RDF/RDFS entailment semantics as described here

Note: haven't implemented rules that produce exceptions; waiting to determine the best way of handling these errors.

Equality Semantics

Property Axiom Semantics

Class Semantics

Class Axiom Semantics

Other

• no datatype semantics for now

Development Notes

To publish new versions of reasonable, tag a commit with the version (e.g. v1.3.2) and push the tag to GitHub. This will execute the publish action which builds an uploads to PyPi.