sansa-stack / sansa-inference

A general Inference API based on two of the most popular Big Data processing engines: Apache Spark and Apache Flink

GitHub

SANSA Inference Layer

Maven Central Build Status License Twitter

Table of Contents

Structure

sansa-inference-common

  • common datastructures
  • rule dependency analysis

sansa-inference-spark

Contains the core Inference API based on Apache Spark.

sansa-inference-flink

Contains the core Inference API based on Apache Flink.

sansa-inference-tests

Contains common test classes and data.

Setup

Prerequisites

  • Maven 3.x
  • Java 8
  • Scala 2.11 (support for Scala 2.12 once Spark moved to Scala 2.12 as well)
  • Apache Spark 2.x
  • Apache Flink 1.x

From source

To install the SANSA Inference API, you need to download it via Git and install it via Maven.

git clone https://github.com/SANSA-Stack/SANSA-Inference.git
cd SANSA-Inference
mvn clean install

Afterwards, you have to add the dependency to your pom.xml

For Apache Spark

<dependency>
  <groupId>net.sansa-stack</groupId>
  <artifactId>sansa-inference-spark_2.11</artifactId>
  <version>VERSION</version>
</dependency>

and for Apache Flink

<dependency>
  <groupId>net.sansa-stack</groupId>
  <artifactId>sansa-inference-flink_2.11</artifactId>
  <version>VERSION</version>
</dependency>

with VERSION beeing the released version you want to use.

Using Maven pre-build artifacts

The latest release is available in Maven Central, thus, you only have to add the following dependency to your pom.xml:

For Apache Spark

<dependency>
  <groupId>net.sansa-stack</groupId>
  <artifactId>sansa-inference-spark_2.11</artifactId>
  <version>0.4.0</version>
</dependency>

and for Apache Flink

<dependency>
  <groupId>net.sansa-stack</groupId>
  <artifactId>sansa-inference-flink_2.11</artifactId>
  <version>0.4.0</version>
</dependency>

Using SBT

Add the following lines to your SBT file:

For Apache Spark add

libraryDependencies += "net.sansa-stack" % "sansa-inference-spark_2.11" % "0.4.0"

and for Apache Flink add

libraryDependencies += "net.sansa-stack" % "sansa-inference-flink_2.11" % "0.4.0"

Using Snapshots

Snapshot version are only avalibale via our custom Maven repository located at http://maven.aksw.org/archiva/repository/snapshots .

Usage

Besides using the Inference API in your application code, we also provide a command line interface with various options that allow for a convenient way to use the core reasoning algorithms:

RDFGraphMaterializer 0.4.0
Usage: RDFGraphMaterializer [options]

  -i, --input <path1>,<path2>,...
                           path to file or directory that contains the input files (in N-Triples format)
  -o, --out <directory>    the output directory
  --properties <property1>,<property2>,...
                           list of properties for which the transitive closure will be computed (used only for profile 'transitive')
  -p, --profile {rdfs | rdfs-simple | owl-horst | transitive}
                           the reasoning profile
  --single-file            write the output to a single file in the output directory
  --sorted                 sorted output of the triples (per file)
  --parallelism <value>    the degree of parallelism, i.e. the number of Spark partitions used in the Spark operations
  --help                   prints this usage text

This can easily be used when submitting the Job to Spark (resp. Flink), e.g. for Spark

/PATH/TO/SPARK/sbin/spark-submit [spark-options] /PATH/TO/INFERENCE-SPARK-DISTRIBUTION/FILE.jar [inference-api-arguments]

and for Flink

/PATH/TO/FLINK/bin/flink run [flink-options] /PATH/TO/INFERENCE-FLINK-DISTRIBUTION/FILE.jar [inference-api-arguments]

In addition, we also provide Shell scripts that wrap the Spark (resp. Flink) deployment and can be used by first setting the environment variable SPARK_HOME (resp. FLINK_HOME) and then calling

/PATH/TO/INFERENCE-DISTRIBUTION/bin/cli [inference-api-arguments]

(Note, that setting Spark (resp. Flink) options isn't supported here and has to be done via the corresponding config files)

Example

RDFGraphMaterializer -i /PATH/TO/FILE/test.nt -o /PATH/TO/TEST_OUTPUT_DIRECTORY/ -p rdfs

will compute the RDFS materialization on the data contained in test.nt and write the inferred RDF graph to the given directory TEST_OUTPUT_DIRECTORY.

Supported Reasoning Profiles

Currently, the following reasoning profiles are supported:

RDFS
RDFS Simple

A fragment of RDFS that covers the most relevant vocabulary, prove that it preserves the original RDFS semantics, and avoids vocabulary and axiomatic information that only serves to reason about the structure of the language itself and not about the data it describes. It is composed of the reserved vocabulary rdfs:subClassOf, rdfs:subPropertyOf, rdf:type, rdfs:domain and rdfs:range.

More details can be found in

Sergio Muñoz, Jorge Pérez, Claudio Gutierrez: Simple and Efficient Minimal RDFS. J. Web Sem. 7(3): 220-234 (2009)

OWL Horst

OWL Horst is a fragment of OWL and was proposed by Herman ter Horst [1] defining an "intentional" version of OWL sometimes also referred to as pD*. It can be materialized using a set of rules that is an extension of the set of RDFS rules. OWL Horst is supposed to be one of the most common OWL flavours for scalable OWL reasoning while bridging the gap between the unfeasible OWL Full and the low expressiveness of RDFS.

[1] Herman J. ter Horst: Completeness, decidability and complexity of entailment for RDF Schema and a semantic extension involving the OWL vocabulary. J. Web Sem. 3(2-3): 79-115 (2005)

How to Contribute

We always welcome new contributors to the project! Please see our contribution guide for more details on how to get started contributing to SANSA.