Examples with TokenStreamToAutomaton org.apache.lucene.analysis.TokenStreamToAutomaton used on opensource projects

Example 1 with TokenStreamToAutomaton

use of org.apache.lucene.analysis.TokenStreamToAutomaton in project elasticsearch by elastic.

the class XAnalyzingSuggester method toAutomaton.

final Automaton toAutomaton(TokenStream ts, final TokenStreamToAutomaton ts2a) throws IOException {
    // Create corresponding automaton: labels are bytes
    // from each analyzed token, with byte 0 used as
    // separator between tokens:
    Automaton automaton = ts2a.toAutomaton(ts);
    automaton = replaceSep(automaton);
    automaton = convertAutomaton(automaton);
    return automaton;
}

Example 2 with TokenStreamToAutomaton

use of org.apache.lucene.analysis.TokenStreamToAutomaton in project elasticsearch by elastic.

the class XAnalyzingSuggester method build.

@Override
public void build(InputIterator iterator) throws IOException {
    String prefix = getClass().getSimpleName();
    Directory tempDir = getTempDir();
    OfflineSorter sorter = new OfflineSorter(tempDir, prefix, new AnalyzingComparator(hasPayloads));
    IndexOutput tempInput = tempDir.createTempOutput(prefix, "input", IOContext.DEFAULT);
    OfflineSorter.ByteSequencesWriter writer = new OfflineSorter.ByteSequencesWriter(tempInput);
    OfflineSorter.ByteSequencesReader reader = null;
    hasPayloads = iterator.hasPayloads();
    BytesRefBuilder scratch = new BytesRefBuilder();
    TokenStreamToAutomaton ts2a = getTokenStreamToAutomaton();
    String tempSortedFileName = null;
    count = 0;
    byte[] buffer = new byte[8];
    try {
        ByteArrayDataOutput output = new ByteArrayDataOutput(buffer);
        for (BytesRef surfaceForm; (surfaceForm = iterator.next()) != null; ) {
            LimitedFiniteStringsIterator finiteStrings = new LimitedFiniteStringsIterator(toAutomaton(surfaceForm, ts2a), maxGraphExpansions);
            for (IntsRef string; (string = finiteStrings.next()) != null; count++) {
                Util.toBytesRef(string, scratch);
                // length of the analyzed text (FST input)
                if (scratch.length() > Short.MAX_VALUE - 2) {
                    throw new IllegalArgumentException("cannot handle analyzed forms > " + (Short.MAX_VALUE - 2) + " in length (got " + scratch.length() + ")");
                }
                short analyzedLength = (short) scratch.length();
                // compute the required length:
                // analyzed sequence + weight (4) + surface + analyzedLength (short)
                int requiredLength = analyzedLength + 4 + surfaceForm.length + 2;
                BytesRef payload;
                if (hasPayloads) {
                    if (surfaceForm.length > (Short.MAX_VALUE - 2)) {
                        throw new IllegalArgumentException("cannot handle surface form > " + (Short.MAX_VALUE - 2) + " in length (got " + surfaceForm.length + ")");
                    }
                    payload = iterator.payload();
                    // payload + surfaceLength (short)
                    requiredLength += payload.length + 2;
                } else {
                    payload = null;
                }
                buffer = ArrayUtil.grow(buffer, requiredLength);
                output.reset(buffer);
                output.writeShort(analyzedLength);
                output.writeBytes(scratch.bytes(), 0, scratch.length());
                output.writeInt(encodeWeight(iterator.weight()));
                if (hasPayloads) {
                    for (int i = 0; i < surfaceForm.length; i++) {
                        if (surfaceForm.bytes[i] == payloadSep) {
                            throw new IllegalArgumentException("surface form cannot contain unit separator character U+001F; this character is reserved");
                        }
                    }
                    output.writeShort((short) surfaceForm.length);
                    output.writeBytes(surfaceForm.bytes, surfaceForm.offset, surfaceForm.length);
                    output.writeBytes(payload.bytes, payload.offset, payload.length);
                } else {
                    output.writeBytes(surfaceForm.bytes, surfaceForm.offset, surfaceForm.length);
                }
                assert output.getPosition() == requiredLength : output.getPosition() + " vs " + requiredLength;
                writer.write(buffer, 0, output.getPosition());
            }
            maxAnalyzedPathsForOneInput = Math.max(maxAnalyzedPathsForOneInput, finiteStrings.size());
        }
        writer.close();
        // Sort all input/output pairs (required by FST.Builder):
        tempSortedFileName = sorter.sort(tempInput.getName());
        // Free disk space:
        tempDir.deleteFile(tempInput.getName());
        reader = new OfflineSorter.ByteSequencesReader(tempDir.openChecksumInput(tempSortedFileName, IOContext.READONCE), prefix);
        PairOutputs<Long, BytesRef> outputs = new PairOutputs<>(PositiveIntOutputs.getSingleton(), ByteSequenceOutputs.getSingleton());
        Builder<Pair<Long, BytesRef>> builder = new Builder<>(FST.INPUT_TYPE.BYTE1, outputs);
        // Build FST:
        BytesRefBuilder previousAnalyzed = null;
        BytesRefBuilder analyzed = new BytesRefBuilder();
        BytesRef surface = new BytesRef();
        IntsRefBuilder scratchInts = new IntsRefBuilder();
        ByteArrayDataInput input = new ByteArrayDataInput();
        // Used to remove duplicate surface forms (but we
        // still index the hightest-weight one).  We clear
        // this when we see a new analyzed form, so it cannot
        // grow unbounded (at most 256 entries):
        Set<BytesRef> seenSurfaceForms = new HashSet<>();
        int dedup = 0;
        while (true) {
            BytesRef bytes = reader.next();
            if (bytes == null) {
                break;
            }
            input.reset(bytes.bytes, bytes.offset, bytes.length);
            short analyzedLength = input.readShort();
            analyzed.grow(analyzedLength + 2);
            input.readBytes(analyzed.bytes(), 0, analyzedLength);
            analyzed.setLength(analyzedLength);
            long cost = input.readInt();
            surface.bytes = bytes.bytes;
            if (hasPayloads) {
                surface.length = input.readShort();
                surface.offset = input.getPosition();
            } else {
                surface.offset = input.getPosition();
                surface.length = bytes.length - surface.offset;
            }
            if (previousAnalyzed == null) {
                previousAnalyzed = new BytesRefBuilder();
                previousAnalyzed.copyBytes(analyzed);
                seenSurfaceForms.add(BytesRef.deepCopyOf(surface));
            } else if (analyzed.get().equals(previousAnalyzed.get())) {
                dedup++;
                if (dedup >= maxSurfaceFormsPerAnalyzedForm) {
                    // dups: skip the rest:
                    continue;
                }
                if (seenSurfaceForms.contains(surface)) {
                    continue;
                }
                seenSurfaceForms.add(BytesRef.deepCopyOf(surface));
            } else {
                dedup = 0;
                previousAnalyzed.copyBytes(analyzed);
                seenSurfaceForms.clear();
                seenSurfaceForms.add(BytesRef.deepCopyOf(surface));
            }
            // TODO: I think we can avoid the extra 2 bytes when
            // there is no dup (dedup==0), but we'd have to fix
            // the exactFirst logic ... which would be sort of
            // hairy because we'd need to special case the two
            // (dup/not dup)...
            // NOTE: must be byte 0 so we sort before whatever
            // is next
            analyzed.append((byte) 0);
            analyzed.append((byte) dedup);
            Util.toIntsRef(analyzed.get(), scratchInts);
            //System.out.println("ADD: " + scratchInts + " -> " + cost + ": " + surface.utf8ToString());
            if (!hasPayloads) {
                builder.add(scratchInts.get(), outputs.newPair(cost, BytesRef.deepCopyOf(surface)));
            } else {
                int payloadOffset = input.getPosition() + surface.length;
                int payloadLength = bytes.length - payloadOffset;
                BytesRef br = new BytesRef(surface.length + 1 + payloadLength);
                System.arraycopy(surface.bytes, surface.offset, br.bytes, 0, surface.length);
                br.bytes[surface.length] = (byte) payloadSep;
                System.arraycopy(bytes.bytes, payloadOffset, br.bytes, surface.length + 1, payloadLength);
                br.length = br.bytes.length;
                builder.add(scratchInts.get(), outputs.newPair(cost, br));
            }
        }
        fst = builder.finish();
    //PrintWriter pw = new PrintWriter("/tmp/out.dot");
    //Util.toDot(fst, pw, true, true);
    //pw.close();
    } finally {
        IOUtils.closeWhileHandlingException(reader, writer);
        IOUtils.deleteFilesIgnoringExceptions(tempDir, tempInput.getName(), tempSortedFileName);
    }
}

Example 3 with TokenStreamToAutomaton

use of org.apache.lucene.analysis.TokenStreamToAutomaton in project elasticsearch by elastic.

the class XAnalyzingSuggester method toFiniteStrings.

// EDIT: Adrien, needed by lookup providers
// NOTE: these XForks are unmaintainable, we need to get rid of them...
public Set<IntsRef> toFiniteStrings(TokenStream stream) throws IOException {
    final TokenStreamToAutomaton ts2a = getTokenStreamToAutomaton();
    Automaton automaton;
    try (TokenStream ts = stream) {
        automaton = toAutomaton(ts, ts2a);
    }
    LimitedFiniteStringsIterator finiteStrings = new LimitedFiniteStringsIterator(automaton, maxGraphExpansions);
    Set<IntsRef> set = new HashSet<>();
    for (IntsRef string = finiteStrings.next(); string != null; string = finiteStrings.next()) {
        set.add(IntsRef.deepCopyOf(string));
    }
    return Collections.unmodifiableSet(set);
}

Example 4 with TokenStreamToAutomaton

use of org.apache.lucene.analysis.TokenStreamToAutomaton in project lucene-solr by apache.

the class AnalyzingSuggester method build.

@Override
public void build(InputIterator iterator) throws IOException {
    if (iterator.hasContexts()) {
        throw new IllegalArgumentException("this suggester doesn't support contexts");
    }
    hasPayloads = iterator.hasPayloads();
    OfflineSorter sorter = new OfflineSorter(tempDir, tempFileNamePrefix, new AnalyzingComparator(hasPayloads));
    IndexOutput tempInput = tempDir.createTempOutput(tempFileNamePrefix, "input", IOContext.DEFAULT);
    OfflineSorter.ByteSequencesWriter writer = new OfflineSorter.ByteSequencesWriter(tempInput);
    OfflineSorter.ByteSequencesReader reader = null;
    BytesRefBuilder scratch = new BytesRefBuilder();
    TokenStreamToAutomaton ts2a = getTokenStreamToAutomaton();
    String tempSortedFileName = null;
    count = 0;
    byte[] buffer = new byte[8];
    try {
        ByteArrayDataOutput output = new ByteArrayDataOutput(buffer);
        for (BytesRef surfaceForm; (surfaceForm = iterator.next()) != null; ) {
            LimitedFiniteStringsIterator finiteStrings = new LimitedFiniteStringsIterator(toAutomaton(surfaceForm, ts2a), maxGraphExpansions);
            for (IntsRef string; (string = finiteStrings.next()) != null; count++) {
                Util.toBytesRef(string, scratch);
                // length of the analyzed text (FST input)
                if (scratch.length() > Short.MAX_VALUE - 2) {
                    throw new IllegalArgumentException("cannot handle analyzed forms > " + (Short.MAX_VALUE - 2) + " in length (got " + scratch.length() + ")");
                }
                short analyzedLength = (short) scratch.length();
                // compute the required length:
                // analyzed sequence + weight (4) + surface + analyzedLength (short)
                int requiredLength = analyzedLength + 4 + surfaceForm.length + 2;
                BytesRef payload;
                if (hasPayloads) {
                    if (surfaceForm.length > (Short.MAX_VALUE - 2)) {
                        throw new IllegalArgumentException("cannot handle surface form > " + (Short.MAX_VALUE - 2) + " in length (got " + surfaceForm.length + ")");
                    }
                    payload = iterator.payload();
                    // payload + surfaceLength (short)
                    requiredLength += payload.length + 2;
                } else {
                    payload = null;
                }
                buffer = ArrayUtil.grow(buffer, requiredLength);
                output.reset(buffer);
                output.writeShort(analyzedLength);
                output.writeBytes(scratch.bytes(), 0, scratch.length());
                output.writeInt(encodeWeight(iterator.weight()));
                if (hasPayloads) {
                    for (int i = 0; i < surfaceForm.length; i++) {
                        if (surfaceForm.bytes[i] == PAYLOAD_SEP) {
                            throw new IllegalArgumentException("surface form cannot contain unit separator character U+001F; this character is reserved");
                        }
                    }
                    output.writeShort((short) surfaceForm.length);
                    output.writeBytes(surfaceForm.bytes, surfaceForm.offset, surfaceForm.length);
                    output.writeBytes(payload.bytes, payload.offset, payload.length);
                } else {
                    output.writeBytes(surfaceForm.bytes, surfaceForm.offset, surfaceForm.length);
                }
                assert output.getPosition() == requiredLength : output.getPosition() + " vs " + requiredLength;
                writer.write(buffer, 0, output.getPosition());
            }
            maxAnalyzedPathsForOneInput = Math.max(maxAnalyzedPathsForOneInput, finiteStrings.size());
        }
        CodecUtil.writeFooter(tempInput);
        writer.close();
        // Sort all input/output pairs (required by FST.Builder):
        tempSortedFileName = sorter.sort(tempInput.getName());
        // Free disk space:
        tempDir.deleteFile(tempInput.getName());
        reader = new OfflineSorter.ByteSequencesReader(tempDir.openChecksumInput(tempSortedFileName, IOContext.READONCE), tempSortedFileName);
        PairOutputs<Long, BytesRef> outputs = new PairOutputs<>(PositiveIntOutputs.getSingleton(), ByteSequenceOutputs.getSingleton());
        Builder<Pair<Long, BytesRef>> builder = new Builder<>(FST.INPUT_TYPE.BYTE1, outputs);
        // Build FST:
        BytesRefBuilder previousAnalyzed = null;
        BytesRefBuilder analyzed = new BytesRefBuilder();
        BytesRef surface = new BytesRef();
        IntsRefBuilder scratchInts = new IntsRefBuilder();
        ByteArrayDataInput input = new ByteArrayDataInput();
        // Used to remove duplicate surface forms (but we
        // still index the hightest-weight one).  We clear
        // this when we see a new analyzed form, so it cannot
        // grow unbounded (at most 256 entries):
        Set<BytesRef> seenSurfaceForms = new HashSet<>();
        int dedup = 0;
        while (true) {
            BytesRef bytes = reader.next();
            if (bytes == null) {
                break;
            }
            input.reset(bytes.bytes, bytes.offset, bytes.length);
            short analyzedLength = input.readShort();
            analyzed.grow(analyzedLength + 2);
            input.readBytes(analyzed.bytes(), 0, analyzedLength);
            analyzed.setLength(analyzedLength);
            long cost = input.readInt();
            surface.bytes = bytes.bytes;
            if (hasPayloads) {
                surface.length = input.readShort();
                surface.offset = input.getPosition();
            } else {
                surface.offset = input.getPosition();
                surface.length = bytes.length - surface.offset;
            }
            if (previousAnalyzed == null) {
                previousAnalyzed = new BytesRefBuilder();
                previousAnalyzed.copyBytes(analyzed.get());
                seenSurfaceForms.add(BytesRef.deepCopyOf(surface));
            } else if (analyzed.get().equals(previousAnalyzed.get())) {
                dedup++;
                if (dedup >= maxSurfaceFormsPerAnalyzedForm) {
                    // dups: skip the rest:
                    continue;
                }
                if (seenSurfaceForms.contains(surface)) {
                    continue;
                }
                seenSurfaceForms.add(BytesRef.deepCopyOf(surface));
            } else {
                dedup = 0;
                previousAnalyzed.copyBytes(analyzed);
                seenSurfaceForms.clear();
                seenSurfaceForms.add(BytesRef.deepCopyOf(surface));
            }
            // TODO: I think we can avoid the extra 2 bytes when
            // there is no dup (dedup==0), but we'd have to fix
            // the exactFirst logic ... which would be sort of
            // hairy because we'd need to special case the two
            // (dup/not dup)...
            // NOTE: must be byte 0 so we sort before whatever
            // is next
            analyzed.append((byte) 0);
            analyzed.append((byte) dedup);
            Util.toIntsRef(analyzed.get(), scratchInts);
            //System.out.println("ADD: " + scratchInts + " -> " + cost + ": " + surface.utf8ToString());
            if (!hasPayloads) {
                builder.add(scratchInts.get(), outputs.newPair(cost, BytesRef.deepCopyOf(surface)));
            } else {
                int payloadOffset = input.getPosition() + surface.length;
                int payloadLength = bytes.length - payloadOffset;
                BytesRef br = new BytesRef(surface.length + 1 + payloadLength);
                System.arraycopy(surface.bytes, surface.offset, br.bytes, 0, surface.length);
                br.bytes[surface.length] = PAYLOAD_SEP;
                System.arraycopy(bytes.bytes, payloadOffset, br.bytes, surface.length + 1, payloadLength);
                br.length = br.bytes.length;
                builder.add(scratchInts.get(), outputs.newPair(cost, br));
            }
        }
        fst = builder.finish();
    //Util.dotToFile(fst, "/tmp/suggest.dot");
    } finally {
        IOUtils.closeWhileHandlingException(reader, writer);
        IOUtils.deleteFilesIgnoringExceptions(tempDir, tempInput.getName(), tempSortedFileName);
    }
}

Example 5 with TokenStreamToAutomaton

use of org.apache.lucene.analysis.TokenStreamToAutomaton in project elasticsearch by elastic.

the class XAnalyzingSuggester method getTokenStreamToAutomaton.

public TokenStreamToAutomaton getTokenStreamToAutomaton() {
    final TokenStreamToAutomaton tsta;
    if (preserveSep) {
        tsta = new EscapingTokenStreamToAutomaton((char) sepLabel);
    } else {
        // When we're not preserving sep, we don't steal 0xff
        // byte, so we don't need to do any escaping:
        tsta = new TokenStreamToAutomaton();
    }
    tsta.setPreservePositionIncrements(preservePositionIncrements);
    return tsta;
}