RDFpro invocation

The command line tool can be invoked using the rdfpro script. Its syntax is:

rdfpro -v | -h | [-V] SPEC

where options -v and -h display respectively the tool version and its online help text, while SPEC is the specification of the RDF processing pipeline that is built and executed by the tool; option -V enables the ‘verbose’ mode where additional debugging information is logged.

Pipeline specification

The pipeline specification SPEC is based on the recursive application of the following rules:

SPEC ::=
    @p args                            instantiates builtin processor @p with argument list args
    @p1 args1 ... @pN argsN            sequence composition of processors @p1 ... @pN with their args
    { @p1 args1 , ... , @pN argsN }S   parallel composition of processors @p1 ... @pN with their args
                                       and their outputs merged using set/multiset operator S

Denoting with Mo(q), Mi(q) the mulitplicity of a quad in the output stream, i-th input stream, the available set/multiset operators S and their semantics are:

  • <no letter> or a – multiset sum of @p1@pN outputs, with Mo(q) = sum(Mi(q)) (very fast);
  • u – set union of @p1@pN outputs, with Mo(q) = max(Mi(q), 1) (same behaviour of processor @unique);
  • U – multiset union of @p1@pN outputs, with Mo(q) = max(Mi(1));
  • i – set intersection of @p1@pN outputs, with Mo(q) = min(Mi(q), 1);
  • I – multiset intersection of @p1@pN outputs, with Mo(q) = min(Mi(q));
  • d – set difference @p1 \ @p2 \ … \ @pN), with Mo(q) = max(0, min(M0(q), 1) - sum_i>0(Mi(q)));
  • D – multiset difference @p1 \ @p2 \ … \ @pN), with Mo(q) = max(0, M0(q) - sum_i>0(Mi(q)));
  • s – symmetric set difference of @p1 and @p2 outputs, i.e. @p1 \ @p2 union @p2 \ @p1 (using set difference and union);
  • S – symmetric multiset difference of @p1 and @p2 outputs, i.e. @p1 \ @p2 union @p2 \ @p1 (using multiset difference and union);
  • n+ – quads with at least n (number) occurrences (no duplicates emitted);
  • n- – quads with at most n (number) occurrences (no duplicates emitted).

Note that set operators different from a are implemented using sorting (sort) and thus are slower than the default a.

As an example, the following command invokes rdfpro in verbose mode with a pipeline that reads a Turtle+gzip file file.ttl.gz, extract TBox and VOID statistics in parallel and writes their set union (u flag) to the RDF/XML file onto.rdf.

rdfpro -V @read file.ttl { @stats , @tbox }u @write onto.rdf

I/O processors

The builtin processors @read, @write, @query and @update allow to move data in and out of RDFpro. We describe them below, reporting both long and short name (e.g., @read and @r) and their arguments.


@read|@r [-b BASE] [-w] URL...

Reads quads from one or more files, injecting them in the input stream to produce the emitted output stream. Multiple files are read in parallel and multiple threads are used to parse files that use a line-oriented RDF syntax (NTriples, NQuads, TQL).

Option -b BASE can be used to specify a base URI for resolving relative URIs in the input files.

Option -w causes BNodes in input files to be rewritten to avoid possible clashes between BNodes in different files or between a parsed BNode and a BNode in the processor input stream. If not specified, the default is to preserve BNodes read from files.

Arguments URL... identify the files to read; local files (absolute or relative paths) as well as file://, http://, https:// URLs can be supplied. For each file, its RDF format and compression scheme are detected based on the extension (e.g., ttl.gz -> gzipped Turtle). This information must be explicitly provided in case the extension is not informative, by prepending the correct extension as .ext: to the URL (e.g., by transforming my_unknown_file to .ttl.gz:my_unknown_file).

The following RDF formats are detected and supported: rdf, rj, jsonld, nt, nq, trix, trig, tql, ttl, n3, brf, geonames. The following compression schemes are detected and supported (provided the corresponding native compression/decompression utility is available): gz, bz2, xz, 7z. Shell expansion can be exploited to list multiple files.


@write|@w [-c NUM] URL...

Writes quads from the input stream to files.

Option -c specifies the number of consecutive quads (from the input stream) to write as a chunk to each file, in case multiple files are specified. This option can impact on the size of produced files, in case compression is used and when quads of the input stream are somehow sorted (e.g., because produced by a call to @unique). In this situation, keeping consecutive (and thus similar) quads together by increasing the value of this option reduces the total size of produced compressed files.

Arguments URL... identify the files to write. Currently, only file:// URLs, possibly given as absolute or relative paths, can be used. If multiple files are specified, quads are allocated to them according to a round-robin strategy that produces files of similar sizes. This behaviour can be used to split a large dataset into smaller and more manageable files. RDF formats and compression schemes are detected from the file extensions, or (similarly to @read) can be explicitly supplied using the notation .ext:filename (e.g., .ttl.gz:filename).

The same RDF formats (except geonames) and compression schemes of @read are supported. The output stream of this processor is the input stream unchanged, thus allowing to chain @write with other downstream processors.


@query [-q QUERY] [-f FILE] [-w] URL

Downloads quads from a SPARQL endpoint using a SPARQL CONSTRUCT or SELECT query.

Option -q QUERY supplies the SPARQL query inline. SELECT queries produce quads based on the bindings of variables s, p, o, c. CONSTRUCT queries return quads in the default graph (i.e., plain triples).

Option -f FILE provides a file containing the SPARQL query to be submitted. This option is mutually exclusive with option -q.

Option -w causes BNodes in downloaded data to be rewritten to avoid possible clashes. The default is to keep them unchanged. Note however that BNodes from a SPARQL query may change from execution to execution depending on the endpoint implementation, as in principle they are scoped locally to the query.

Argument URL specifies the URL of the SPARQL endpoint.

Downloaded quads are emitted in the output stream together with quads from the input stream.


@update [-s SIZE] URL

Uploads quads in the input stream to a SPARQL endpoint using SPARQL Update INSERT DATA calls.

Option -s SIZE specifies the maximum size of a chunk of data uploaded with a SPARQL Update call (default 1024 quads).

Argument URL specifies the URL of the SPARQL endpoint.

The output stream of this processor is the input stream unchanged, thus allowing to chain @update with other downstream processors.

Data transformation processors

Processors @transform, @groovy, @rdfs, rules, mapreduce, @smush, @unique implement different forms of data transformation.


@transform|@t [EXP]

Discards and/or replaces quads based on supplied expression.

The expression can be either a script (language: script file / script expression) or a set of simple matching/rewrite rules detailed below.

Given X a quad component (possible values: s, p, o, c), the string contains three types of rules:

  • +X value list – quad is dropped if X does not belong to value list;
  • -X value list – quad is dropped if X belongs to value list;
  • =X value – quad component X is replaced with value (evaluated after filters).

Note: for a given component X, only a rule +X or -X can appear. If you have more than one of such rules, consider adding multiple @transform processors to your pipeline. Values must be encoded in Turtle. The following wildcard values are supported:

  • <*> – any URI;
  • _:* – any BNode;
  • * – any literal;
  • *@* – any literal with a language;
  • *@xyz – any literal with language xyz;
  • *^^* – any typed literal;
  • *^^<uri> – any literal with datatype <uri>;
  • *^^ns:uri – any literal with datatype ns:uri;
  • *^^ns:* – any typed literal with datatype prefixed with ns:;
  • ns:* – any URI prefixed with ns:;
  • <ns*> – any URI with namespace URI ns.

Examples. We report some concrete examples of how to use the filtering mechanism:

  • @transform '-p foaf:name' – removes all foaf:name quads;
  • @transform '+p rdf:type +s <*> +o <*>' – keeps only rdf:type quads whose subject and object are URIs;
  • @transform '+p rdf:type =c <http://example.org/mygraph>' – extracts rdf:type quads, placing them in named graph <http://example.org/mygraph>.


@groovy [-p] SCRIPT [ARG...]

Transform the stream using a user-supplied Groovy SCRIPT. Extra arguments ARG... are passed to the script. Option -p enables script pooling, i.e., the execution of multiple script instances in parallel each one on a subset of the input stream; pooling is faster but makes meaningless the use of global variables in the script (there is no such thing as a global state).

The script is executed for each quad (preserving script variables among invocations) passing it the following modifiable variable (note: changing a variable implies a modification of the current quad):

  • q - the current quad being transformed
  • s, p, o, c the quad subject, predicate, object and graph (context)
  • t the type of an rdf:type quad (otherwise undefined)
  • l and d the literal label and datatype (undefined if object is not a literal)

In alternative, the script may provide a function handle(quad) that is called for each quad in place of executing the whole script. Independently of the use of handle(quad), the following three callback functions can be defined by the script:

  • init(args) - called when the script is initialized with the ARG... arguments supplied to the @groovy processor
  • start(pass) - called each time a pass on input data is started, supplying the 0-based integer index of the pass
  • end(pass) - called each time a pass on input data is completed, supplying the 0-based integer index of the pass

The body of the script may use any Groovy construct or library function (including Java libraries that are accessible in Groovy). In addition, scripts may use the syntax prefix:name and <uri_string> to enter respectively QNames and full URIs (resolved to Sesame URI constants). Groovy implementation of SPARQL 1.1 functions are available to the script (just enter their name in lowercase) as well as the following special functions

  • emit() - emits the input quad (possibly changed by operating on variables q, s, p, o, c, t, l, d)
  • emitIf(condition) - emits the input quad only if condition evaluates to true
  • emitIfNot(condition) - emits the input quad only if condition evaluates to false
  • emit(quad) - emits the supplied quad, passed as a Sesame Statement object
  • emit(s, p, o, c) - emits the quad formed by the supplied subject, predicate, object and context (any value can be passed for them - proper conversion will be done by RDFpro)
  • error(message) - halts the script execution with the supplied error message
  • error(message, throwable) - halts the script execution with the supplied error message and exception object
  • log(message) - logs the specified message

Important: input quads are never propagated automatically. Emission of quads in the output stream occurs only by invoking one of the emitXXX(...) methods!

Finally, note that the boolean flag __rdfpro__ is set to true when the script is run inside RDFpro (to distinguish from other forms of script invocation, e.g. from the command line).


@rdfs [-e RULES] [-d] [-t] [-C | -c URI] [-b BASE] [-w] [URL...]

Emits the RDFS deductive closure of input quads. One or more TBox files are loaded and their RDFS closure is computed and (possibly) emitted first. Next, the domain, range, sub-class and sub-property axioms from the TBox are used to do inference on input quads one at a time, placing inferences in the same graph of the input quad.

Option -e RULES specifies the RDFS rules to exclude and accepts a comma-separated list of rule names. Available rules are rdfd2, rdfs1, rdfs2, rdfs3, rdfs4a, rdfs4b, rdfs5, rdfs6, rdfs7, rdfs8, rdfs9, rdfs10, rdfs11, rdfs12, rdfs13. Rules rdfs4a, rdfs4b and rdfs8 generates a lot of quads of the form <x rdf:type rdfs:Resource>. Exclude these rules to reduce the volume of generated data.

Option -d causes simple OWL axioms in the TBox to be decomposed to their corresponding RDFS axioms, where possible, so that they can affect RDFS reasoning (e.g., owl:equivalentClass can be expressed in terms of rdfs:subClassOf).

Option -t causes uninformative <x rdf:type _:b> statements, with _:b a BNode, to be dropped (default: keep).

Options -C and -c URI control the graph where the TBox closure is emitted. This graph is the default (unnamed) graph if option -C is specified, otherwise the URI given by -c is used as the targed graph. If none of these options is specified, the TBox closure is not emitted.

Option -b BASE specifies the base URI to be used for resolving relative URIs in the TBox file.

Option -w can be used to rewrite BNodes in TBox files, similarly to the corresponding option in @read.

Arguments URL... identify the TBox files to be read. The same restrictions and explicit extension prefix supported by @read can be used.

Note. The algorighm used by @rdfs avoids expensive join operations and works with arbitrarily large datasets, provided that their TBox fits into memory. The result represents the complete RDFS closure if the TBOX: (i) contains all the rdfs:domain, rdfs:range, rdfs:subClassOf or rdfs:subPropertyOf axioms in the input stream; and (ii) it contains no quad matching patterns:

  • X rdfs:subPropertyOf {rdfs:domain|rdfs:range|rdfs:subPropertyOf|rdfs:subClassOf}
  • X {rdf:type|rdfs:domain|rdfs:range|rdfs:subClassOf} rdfs:ContainerMembershipProperty
  • X {rdf:type|rdfs:domain|rdfs:range|rdfs:subClassOf} rdfs:Datatype


@rules [-r RULESETS] [-B BINDINGS] [-p MODE] [-g MODE] [-G URI] [-t] [-u] [-C | -c URI] [-b URI] [-w] URL...

Emit the closure of input quads using the specified RULESETS (comma-separated list), possibly pre-processing rules based on supplied TBox data provided by arguments URL....

Option -r is a comma-separated list of ruleset names. Names of local files as well URLs can be used. In addition, names rdfs and owl2rl refer to builtin rulesets for RDFS and OWL 2 RL inference.

Option -B is a comma-separated list of variable=value bindings (values are URIs or literals). These bindings are used to replace corresponding variables in the rules, thus customizing them.

Option -p specifies if and how to partition input quads. Acceptable values are none for no partitioning (default); entity for partitioning data by entities, i.e., subjects and URI/BNode objects (note: if a quad <s,p,o,c> has a URI or BNode object o, it will be assigned to both partitions for s and for o); and graph for partitioning data by named graph.

Option -g specifies if and how to modify the input ruleset to take into accounts named graphs. Default value none corresponds to no modifications; global means that rules are modified so to match premises in any graph and emit consequences in a global graph specified using option -G; value separate means that rules will match premises and place consequence exactly in the same graph, with the effect that inference is done separately on each graph; value star is a combination of global and separate, meaning that inference is done separately per graph but premises can also match quads in a special global graph supplied with option g (the closure of this global graph is placed in that graph itself).

Option -t causes uninformative <x rdf:type _:b> statements, with _:b a BNode, to be dropped (default: keep).

Option -u enforces the emission of unique statements and can be more efficient than performing a subsequent invocation to processor @unique. Without it, the rule engine might return duplicates if this allows it to operate faster.

Options -C and -c URI control the graph where the closure of TBox data is emitted. Option -C causes the closure to be emitted as is, using the same graphs computed in the closure. Option -c causes the closure to be emitted in a specific named graph. If none of these options is specified, the closure of TBox data is not emitted.

Option -b BASE specifies the base URI to be used for resolving relative URIs in TBox data. Option -w can be used to rewrite BNodes in TBox data, similarly to the corresponding option in @read.

Arguments URL... identify the TBox files to be read. Typically these files supply TBox quads, but this is not a restriction. In case TBox data is supplied, the processor will compute its closure using the chosen rulesets. Then, rules will be pre-processed with respect to this closure, exploding the TBox parts of their bodies (based on rr:MetaVocabularyTerm declarations in the rulesets) and deriving more specific rules that are used to perform inference on ABox data arriving from the input stream.


@mapreduce [-b PRED] [-r REDUCER] [-e PRED] [-a PRED] [-u] MAPPER...

Performs a MapReduce computation, using an in-process, multi-thread implementation of MapReduce optimized using RDFpro sort implementation optimized for dealing with RDF data.

Arguments MAPPER... specify the mappers to apply to quads of the input stream. Each mapper could be a script implementing the eu.fbk.rdfpro.Mapper interface, either supplied inline or in a file. Other values can be e, to select a mapper that partitions quads by entity (i.e., subject and URI/BNode object), and a string satisfying regex [spoc]+ for a mapper that returns the hash of the selected subject (s), predicate (p), object (o) and context (c) quad components.

Option -r specifies the reducer. The reducer should be a script implementing the interface eu.fbk.rdfpro.Reducer. If omitted, the default identity reducer that emits the partition unchanged is used.

Option -b specifies a bypass predicate applied to input quads; quads matching the predicate are emitted in output without being processed by the MapReduce machinery.

Option -e specifies an existential filtering predicate that must be satisfied by at least a partition quad for the partition to be reduced (otherwise, the partition is silently discarded). This option is useful to achieve effects such as: partition by entity and return quads of entities of type / with property X.

Option -a specifies a universal filtering predicate that must be satisfied by all the quads of a partition in order for the partition to be reduced (otherwise, it is discarded).

Option -u requires the processor to deduplicate quads of each partition before processing them (this can be obtained for free and should be preferred to invoke @unique).


@smush NAMESPACE...

Performs smushing, i.e., identifies owl:sameAs equivalence classes and, for each of them, selects a URI as the ‘canonical URI’ for the class which replaces other alias URIs in input quads.

Arguments NAMESPACE... provides the ranked list of namespace URIs or prefixes (resolved based on prefix.cc used to select the canonical URI. At least a namespace must be supplied.

Aliases are not discarded but are emitted using owl:sameAs quads that link them to canonical URIs.


@unique|@u [-m]

Discards duplicates in the input stream, using external sorting.

Option -m causes quads with the same s,p,o components but different graphs to be merged in a new graph that represents the ‘fusion’ of the source graphs (if more than one, otherwise the unique source graph is reused). The fusion graph is described with (i.e., it is the subject of) all the quads that describe the associated source graphs.

Other processors

The remaining processors @stats, @tbox serve various extraction tasks, while @prefix can be used to improve the serialization of produced RDF.


@stats [-n NAMESPACE] [-p URI] [-c URI] [-t NUM] [-o]

Emits VOID structural statistics for input quads. A VOID dataset is associated to the whole input and to each set of graphs associated to the same ‘source’ URI with a configurable property in a configurable graph. Class and property partitions are then generated for each of these datasets. In addition to standard VOID terms, the processor emits additional quads based on an extension vocabulary to express the number of TBox, ABox, rdf:type and owl:sameAs quads, the average number of properties per entity and informative labels and examples for each TBox term, which are then viewable in tools such as Protégé.

Option -n NAMESPACE specifies the namespace where to put generated URIs for VOID dataset (default stats:).

Option -p URI specifies the property linking graphs to their source URI. If not supplied, graph-source link will not be read.

Option -c URI specifies the graph where to look for graph-source links. If not supplied, links will be gathered from any graph.

Option -t NUM causes the emission of VOID statistics only for concepts having at least NUM instances or quads. This option can be used to reduce the amount of data generated by @stats, especially if the output must be visualized in tools such as Protégé.

Option -o enables the computation of void:classes and void:properties, which is memory-intensive (computation may fail if thousands or more of distinct properties are used in the data).

Internally, @stats makes use of the sort utility to (conceptually) sort the quad stream twice: first based on the subject to group quads about the same entity and compute entity-based and distinct subjects statistics; then based on the object to compute distinct objects statistics. Therefore, computing VOID statistics is a quite slow operation.



Filters the input stream by emitting only quads belonging to RDFS or OWL TBox axioms (no check is done on the graph component). Note that OWL TBox axioms are not used for inference by RDFpro (only RDFS is supported).


@prefix|@p [-f FILE]

Adds prefix-to-namespace bindings to data in the input stream so to enable writing more compact and readable files.

Option -f FILE specifies the file providing available bindings, consisting of multiple namespace prefix1 prefix2 ... lines each associating a namespace URI to one or more prefixes. If not specified, the default is to use bindings from prefix.cc.

Back to top

Last Published: 2016/07/22.

Reflow Maven skin by Andrius Velykis.

Data and Knowledge Management tools