Examples with FeatureSet com.yahoo.document.predicate.FeatureSet used on opensource projects

Example 1 with FeatureSet

use of com.yahoo.document.predicate.FeatureSet in project vespa by vespa-engine.

the class PredicateTreeAnnotator method assignIntervalLabels.

/**
 * Visits the predicate tree in depth-first order and assigns intervals for features in
 * {@link com.yahoo.document.predicate.FeatureSet} and {@link com.yahoo.document.predicate.FeatureRange}.
 */
private static void assignIntervalLabels(Predicate predicate, int begin, int end, boolean isNegated, AnnotatorContext context) {
    // Otherwise, conjunctions and disjunctions must be switched if negated (De Morgan's law).
    if (predicate instanceof Conjunction) {
        List<Predicate> children = ((Conjunction) predicate).getOperands();
        int current = begin;
        for (int i = 0; i < children.size(); i++) {
            Predicate child = children.get(i);
            int subTreeSize = context.subTreeSizes.get(child);
            if (i == children.size() - 1) {
                // Last child (and sometimes the only one)
                assignIntervalLabels(child, current, end, isNegated, context);
            // No need to update/touch current since this is the last child.
            } else if (i == 0) {
                // First child
                int next = context.leftNodeLeaves + subTreeSize + 1;
                assignIntervalLabels(child, current, next - 1, isNegated, context);
                current = next;
            } else {
                // Middle children
                int next = current + subTreeSize;
                assignIntervalLabels(child, current, next - 1, isNegated, context);
                current = next;
            }
        }
    } else if (predicate instanceof FeatureConjunction) {
        // Register FeatureConjunction as it was a FeatureSet with a single child.
        // Note: FeatureConjunction should never be negated as AndOrSimplifier will push negations down to
        // the leafs (FeatureSets).
        int zStarEnd = isNegated ? calculateZStarIntervalEnd(end, context) : end;
        IndexableFeatureConjunction indexable = new IndexableFeatureConjunction((FeatureConjunction) predicate);
        int interval = Interval.fromBoundaries(begin, zStarEnd);
        context.featureConjunctions.computeIfAbsent(indexable, (k) -> new ArrayList<>()).add(interval);
        if (isNegated) {
            registerZStarInterval(begin, end, zStarEnd, context);
        }
        context.leftNodeLeaves += 1;
    } else if (predicate instanceof Disjunction) {
        // the values will be same as that of the parent OR node
        for (Predicate child : ((Disjunction) predicate).getOperands()) {
            assignIntervalLabels(child, begin, end, isNegated, context);
        }
    } else if (predicate instanceof FeatureSet) {
        FeatureSet featureSet = (FeatureSet) predicate;
        int zStarEnd = isNegated ? calculateZStarIntervalEnd(end, context) : end;
        for (String value : featureSet.getValues()) {
            long featureHash = Feature.createHash(featureSet.getKey(), value);
            int interval = Interval.fromBoundaries(begin, zStarEnd);
            registerFeatureInterval(featureHash, interval, context.intervals);
        }
        if (isNegated) {
            registerZStarInterval(begin, end, zStarEnd, context);
        }
        context.leftNodeLeaves += 1;
    } else if (predicate instanceof Negation) {
        assignIntervalLabels(((Negation) predicate).getOperand(), begin, end, !isNegated, context);
    } else if (predicate instanceof FeatureRange) {
        FeatureRange featureRange = (FeatureRange) predicate;
        int zStarEnd = isNegated ? calculateZStarIntervalEnd(end, context) : end;
        int interval = Interval.fromBoundaries(begin, zStarEnd);
        for (RangePartition partition : featureRange.getPartitions()) {
            long featureHash = PredicateHash.hash64(partition.getLabel());
            registerFeatureInterval(featureHash, interval, context.intervals);
        }
        for (RangeEdgePartition edgePartition : featureRange.getEdgePartitions()) {
            long featureHash = PredicateHash.hash64(edgePartition.getLabel());
            IntervalWithBounds intervalWithBounds = new IntervalWithBounds(interval, (int) edgePartition.encodeBounds());
            registerFeatureInterval(featureHash, intervalWithBounds, context.intervalsWithBounds);
        }
        if (isNegated) {
            registerZStarInterval(begin, end, zStarEnd, context);
        }
        context.leftNodeLeaves += 1;
    } else {
        throw new UnsupportedOperationException("Cannot handle predicate of type " + predicate.getClass().getSimpleName());
    }
}

Example 2 with FeatureSet

use of com.yahoo.document.predicate.FeatureSet in project vespa by vespa-engine.

the class PredicateTreeAnalyzer method aggregatePredicateStatistics.

// First analysis pass. Traverses tree in depth-first order. Determines the sub-tree sizes and counts the occurrences
// of each feature (used by min-feature calculation in second pass).
// Returns the size of the analyzed subtree.
private static int aggregatePredicateStatistics(Predicate predicate, boolean isNegated, AnalyzerContext context) {
    if (predicate instanceof Negation) {
        return aggregatePredicateStatistics(((Negation) predicate).getOperand(), !isNegated, context);
    } else if (predicate instanceof Conjunction) {
        return ((Conjunction) predicate).getOperands().stream().mapToInt(child -> {
            int size = aggregatePredicateStatistics(child, isNegated, context);
            context.subTreeSizes.put(child, size);
            return size;
        }).sum();
    } else if (predicate instanceof FeatureConjunction) {
        if (isNegated) {
            context.hasNegationPredicate = true;
            return 2;
        }
        // Count the number of identical feature conjunctions - use the id from IndexableFeatureConjunction as key
        IndexableFeatureConjunction ifc = new IndexableFeatureConjunction((FeatureConjunction) predicate);
        incrementOccurrence(context.conjunctionOccurrences, ifc.id);
        // Handled as leaf in interval algorithm - count a single child
        return 1;
    } else if (predicate instanceof Disjunction) {
        return ((Disjunction) predicate).getOperands().stream().mapToInt(child -> aggregatePredicateStatistics(child, isNegated, context)).sum();
    } else if (predicate instanceof FeatureSet) {
        if (isNegated) {
            context.hasNegationPredicate = true;
            return 2;
        } else {
            FeatureSet featureSet = (FeatureSet) predicate;
            for (String value : featureSet.getValues()) {
                incrementOccurrence(context.featureOccurrences, Feature.createHash(featureSet.getKey(), value));
            }
            return 1;
        }
    } else if (predicate instanceof FeatureRange) {
        if (isNegated) {
            context.hasNegationPredicate = true;
            return 2;
        } else {
            incrementOccurrence(context.featureOccurrences, PredicateHash.hash64(((FeatureRange) predicate).getKey()));
            return 1;
        }
    } else {
        throw new UnsupportedOperationException("Cannot handle predicate of type " + predicate.getClass().getSimpleName());
    }
}

Example 3 with FeatureSet

use of com.yahoo.document.predicate.FeatureSet in project vespa by vespa-engine.

the class PredicateFieldValueSerializationTestCase method requireThatPredicateDeserializationMatchesCpp.

@Test
public void requireThatPredicateDeserializationMatchesCpp() throws IOException {
    assertDeserialize("foo_in_bar_and_baz_in_cox", new Conjunction(new FeatureSet("foo", "bar"), new FeatureSet("baz", "cox")));
    assertDeserialize("foo_in_bar_or_baz_in_cox", new Disjunction(new FeatureSet("foo", "bar"), new FeatureSet("baz", "cox")));
    assertDeserialize("foo_in_6_9", new FeatureRange("foo", 6L, 9L));
    assertDeserialize("foo_in_6_x", new FeatureRange("foo", 6L, null));
    assertDeserialize("foo_in_x_9", new FeatureRange("foo", null, 9L));
    assertDeserialize("foo_in_x_x", new FeatureRange("foo", null, null));
    assertDeserialize("foo_in_x", new FeatureSet("foo"));
    assertDeserialize("foo_in_bar", new FeatureSet("foo", "bar"));
    assertDeserialize("foo_in_bar_baz", new FeatureSet("foo", "bar", "baz"));
    assertDeserialize("not_foo_in_bar", new Negation(new FeatureSet("foo", "bar")));
    assertDeserialize("true", new BooleanPredicate(true));
    assertDeserialize("false", new BooleanPredicate(false));
}

Example 4 with FeatureSet

use of com.yahoo.document.predicate.FeatureSet in project vespa by vespa-engine.

the class PredicateFieldValueSerializationTestCase method requireThatPredicateFieldValuesAreDeserialized.

@Test
public void requireThatPredicateFieldValuesAreDeserialized() {
    Document prevDocument = docFactory.createDocument();
    PredicateFieldValue prevPredicate = new PredicateFieldValue(new Conjunction(new FeatureSet("foo", "bar"), new FeatureRange("baz", 6L, 9L)));
    prevDocument.setFieldValue(PREDICATE_FIELD, prevPredicate);
    byte[] buf = serializeDocument(prevDocument);
    Document nextDocument = deserializeDocument(buf, docFactory);
    assertEquals(prevDocument, nextDocument);
    assertEquals(prevPredicate, nextDocument.getFieldValue(PREDICATE_FIELD));
}

Example 5 with FeatureSet

use of com.yahoo.document.predicate.FeatureSet in project vespa by vespa-engine.

the class PredicateFieldValueTest method requireThatXmlOutputIsPredicateLanguage.

@Test
public void requireThatXmlOutputIsPredicateLanguage() {
    Predicate predicate = new FeatureSet("key", "valueA", "valueB");
    XmlStream expected = new XmlStream();
    expected.beginTag("tag");
    expected.addContent(predicate.toString());
    expected.endTag();
    assertEquals(expected.toString(), printXml("tag", new PredicateFieldValue(predicate)));
}