/**
* Describe the vocabularies which are in use in the KB based on the predicate partition
* statistics.
*
* @param predicateParitionCounts The predicate partition statistics.
*/
protected void describeVocabularies(final IVCount[] predicatePartitionCounts) {
// Find the distinct vocabularies in use.
final Set<String> namespaces = new LinkedHashSet<String>();
{
// property partitions.
for (IVCount tmp : predicatePartitionCounts) {
final URI p = (URI) tmp.getValue();
String namespace = p.getNamespace();
if (namespace.endsWith("#")) {
// Strip trailing '#' per VoID specification.
namespace = namespace.substring(0, namespace.length() - 1);
}
namespaces.add(namespace);
}
}
// Sort into dictionary order.
final String[] a = namespaces.toArray(new String[namespaces.size()]);
Arrays.sort(a);
for (String namespace : a) {
g.add(aDataset, VoidVocabularyDecl.vocabulary, f.createURI(namespace));
}
}
/**
* Describe a named or default graph.
*
* @param graph The named graph.
* @param predicatePartitionCounts The predicate partition statistics for that graph.
* @param classPartitionCounts The class partition statistics for that graph.
*/
protected void describeGraph(
final Resource graph,
final IVCount[] predicatePartitionCounts,
final IVCount[] classPartitionCounts) {
// The graph is a Graph.
g.add(graph, RDF.TYPE, SD.Graph);
// #of triples in the default graph
g.add(graph, VoidVocabularyDecl.triples, f.createLiteral(tripleStore.getStatementCount()));
// #of entities in the default graph.
g.add(graph, VoidVocabularyDecl.entities, f.createLiteral(tripleStore.getURICount()));
// #of distinct predicates in the default graph.
g.add(graph, VoidVocabularyDecl.properties, f.createLiteral(predicatePartitionCounts.length));
// #of distinct classes in the default graph.
g.add(graph, VoidVocabularyDecl.classes, f.createLiteral(classPartitionCounts.length));
// property partition statistics.
for (IVCount tmp : predicatePartitionCounts) {
final BNode propertyPartition = f.createBNode();
final URI p = (URI) tmp.getValue();
g.add(graph, VoidVocabularyDecl.propertyPartition, propertyPartition);
g.add(propertyPartition, VoidVocabularyDecl.property, p);
g.add(propertyPartition, VoidVocabularyDecl.triples, f.createLiteral(tmp.count));
}
// class partition statistics.
{
// per class partition statistics.
for (IVCount tmp : classPartitionCounts) {
final BNode classPartition = f.createBNode();
final BigdataValue cls = tmp.getValue();
g.add(graph, VoidVocabularyDecl.classPartition, classPartition);
g.add(classPartition, VoidVocabularyDecl.class_, cls);
g.add(classPartition, VoidVocabularyDecl.triples, f.createLiteral(tmp.count));
}
} // end class partition statistics.
}
/**
* Return the class partition statistics for the named graph.
*
* @param kb The KB instance.
* @param civ The {@link IV} of a named graph (required).
* @return The class partition statistics for that named graph. Only class partitions which are
* non-empty are returned.
*/
protected static IVCount[] classUsage(
final AbstractTripleStore kb, final IV<?, ?> civ, final IVCount[] classPartitionCounts) {
final SPORelation r = kb.getSPORelation();
final boolean quads = kb.isQuads();
if (!quads) {
// Named graph only valid in quads mode.
throw new IllegalArgumentException();
}
// Resolve IV for rdf:type
final BigdataURI rdfType = kb.getValueFactory().asValue(RDF.TYPE);
kb.getLexiconRelation()
.addTerms(new BigdataValue[] {rdfType}, 1 /* numTerms */, true /* readOnly */);
if (rdfType.getIV() == null) {
// No rdf:type assertions since rdf:type is unknown term.
return new IVCount[0];
}
// The non-zero counts.
final List<IVCount> counts = new LinkedList<IVCount>();
// Check the known non-empty predicate partitions.
for (IVCount in : classPartitionCounts) {
final long n =
r.getAccessPath(null, rdfType.getIV() /* p */, in.iv /* o */, civ)
.rangeCount(false /* exact */);
if (n == 0) continue;
final IVCount out = new IVCount(in.iv, n);
out.setValue(in.getValue());
counts.add(out);
}
final IVCount[] a = counts.toArray(new IVCount[counts.size()]);
// Order by descending count.
Arrays.sort(a);
return a;
}
/**
* Return the predicate partition statistics for the named graph.
*
* @param kb The KB instance.
* @param civ The {@link IV} of a named graph (required).
* @return The predicate partition statistics for that named graph. Only predicate partitions
* which are non-empty are returned.
*/
protected static IVCount[] predicateUsage(
final AbstractTripleStore kb, final IV<?, ?> civ, final IVCount[] predicatePartitionCounts) {
final SPORelation r = kb.getSPORelation();
final boolean quads = kb.isQuads();
if (!quads) {
// Named graph only valid in quads mode.
throw new IllegalArgumentException();
}
// The non-zero counts.
final List<IVCount> counts = new LinkedList<IVCount>();
// Check the known non-empty predicate partitions.
for (IVCount in : predicatePartitionCounts) {
final long n = r.getAccessPath(null, in.iv, null, civ).rangeCount(false /* exact */);
if (n == 0) continue;
final IVCount out = new IVCount(in.iv, n);
out.setValue(in.getValue());
counts.add(out);
}
final IVCount[] a = counts.toArray(new IVCount[counts.size()]);
// Order by descending count.
Arrays.sort(a);
return a;
}
/**
* Return an efficient statistical summary for the class partitions. The SPARQL query for this is
*
* <pre>
* SELECT ?class (COUNT(?s) AS ?count ) { ?s a ?class } GROUP BY ?class ORDER BY ?count
* </pre>
*
* However, it is much efficient to scan POS for
*
* <pre>
* rdf:type ?o ?s
* </pre>
*
* and report the range count of
*
* <pre>
* rdf:type ?o ?s
* </pre>
*
* for each distinct value of <code>?o</code>.
*
* @param kb The KB instance.
* @return The class usage statistics.
*/
protected static IVCount[] classUsage(final AbstractTripleStore kb) {
final SPORelation r = kb.getSPORelation();
if (r.oneAccessPath) {
// The necessary index (POS or POCS) does not exist.
throw new UnsupportedOperationException();
}
final boolean quads = kb.isQuads();
final SPOKeyOrder keyOrder = quads ? SPOKeyOrder.POCS : SPOKeyOrder.POS;
// Resolve IV for rdf:type
final BigdataURI rdfType = kb.getValueFactory().asValue(RDF.TYPE);
kb.getLexiconRelation()
.addTerms(new BigdataValue[] {rdfType}, 1 /* numTerms */, true /* readOnly */);
if (rdfType.getIV() == null) {
// No rdf:type assertions since rdf:type is unknown term.
return new IVCount[0];
}
// visit distinct term identifiers for the rdf:type predicate.
@SuppressWarnings("rawtypes")
final IChunkedIterator<IV> itr =
r.distinctMultiTermScan(keyOrder, new IV[] {rdfType.getIV()} /* knownTerms */);
// resolve term identifiers to terms efficiently during iteration.
final BigdataValueIterator itr2 = new BigdataValueIteratorImpl(kb /* resolveTerms */, itr);
try {
final Set<IV<?, ?>> ivs = new LinkedHashSet<IV<?, ?>>();
final Map<IV<?, ?>, IVCount> counts = new LinkedHashMap<IV<?, ?>, IVCount>();
while (itr2.hasNext()) {
final BigdataValue term = itr2.next();
final IV<?, ?> iv = term.getIV();
final long n =
r.getAccessPath(null, rdfType.getIV() /* p */, iv /* o */, null)
.rangeCount(false /* exact */);
ivs.add(iv);
counts.put(iv, new IVCount(iv, n));
}
// Batch resolve IVs to Values
final Map<IV<?, ?>, BigdataValue> x = kb.getLexiconRelation().getTerms(ivs);
for (Map.Entry<IV<?, ?>, BigdataValue> e : x.entrySet()) {
final IVCount count = counts.get(e.getKey());
count.setValue(e.getValue());
}
final IVCount[] a = counts.values().toArray(new IVCount[counts.size()]);
// Order by descending count.
Arrays.sort(a);
return a;
} finally {
itr2.close();
}
}
/**
* Return an array of the distinct predicates in the KB ordered by their descending frequency of
* use. The {@link IV}s in the returned array will have been resolved to the corresponding {@link
* BigdataURI}s which can be accessed using {@link IV#getValue()}.
*
* @param kb The KB instance.
*/
protected static IVCount[] predicateUsage(final AbstractTripleStore kb) {
final SPORelation r = kb.getSPORelation();
if (r.oneAccessPath) {
// The necessary index (POS or POCS) does not exist.
throw new UnsupportedOperationException();
}
final boolean quads = kb.isQuads();
// the index to use for distinct predicate scan.
final SPOKeyOrder keyOrder = quads ? SPOKeyOrder.POCS : SPOKeyOrder.POS;
// visit distinct term identifiers for predicate position on that index.
@SuppressWarnings("rawtypes")
final IChunkedIterator<IV> itr = r.distinctTermScan(keyOrder);
// resolve term identifiers to terms efficiently during iteration.
final BigdataValueIterator itr2 = new BigdataValueIteratorImpl(kb /* resolveTerms */, itr);
try {
final Set<IV<?, ?>> ivs = new LinkedHashSet<IV<?, ?>>();
final Map<IV<?, ?>, IVCount> counts = new LinkedHashMap<IV<?, ?>, IVCount>();
while (itr2.hasNext()) {
final BigdataValue term = itr2.next();
final IV<?, ?> iv = term.getIV();
final long n = r.getAccessPath(null, iv, null, null).rangeCount(false /* exact */);
ivs.add(iv);
counts.put(iv, new IVCount(iv, n));
}
// Batch resolve IVs to Values
final Map<IV<?, ?>, BigdataValue> x = kb.getLexiconRelation().getTerms(ivs);
for (Map.Entry<IV<?, ?>, BigdataValue> e : x.entrySet()) {
final IVCount count = counts.get(e.getKey());
count.setValue(e.getValue());
}
final IVCount[] a = counts.values().toArray(new IVCount[counts.size()]);
// Order by descending count.
Arrays.sort(a);
return a;
} finally {
itr2.close();
}
}
