Examples with BlockingExecutor zemberek.core.concurrency.BlockingExecutor used on opensource projects

Example 1 with BlockingExecutor

use of zemberek.core.concurrency.BlockingExecutor in project zemberek-nlp by ahmetaa.

the class ProcessNormalizationCorpus method process.

void process(BlockTextLoader corpusProvider, int threadCount, Path outRoot) throws Exception {
    ExecutorService service = new BlockingExecutor(threadCount);
    AtomicInteger c = new AtomicInteger(0);
    for (TextChunk chunk : corpusProvider) {
        service.submit(() -> {
            List<String> sentences = TextCleaner.cleanAndExtractSentences(chunk.getData());
            sentences = sentences.stream().map(s -> normalizer.preProcess(s)).collect(Collectors.toList());
            Path p = outRoot.resolve(String.valueOf(c.getAndIncrement()));
            try {
                Files.write(p, sentences, StandardCharsets.UTF_8);
            } catch (IOException e) {
                e.printStackTrace();
            }
            Log.info(c.get() * BLOCK_SIZE + " Lines processed.");
        });
    }
    service.shutdown();
    service.awaitTermination(1, TimeUnit.DAYS);
}

Example 2 with BlockingExecutor

use of zemberek.core.concurrency.BlockingExecutor in project zemberek-nlp by ahmetaa.

the class StemDisambiguationExperiment method doit.

private void doit() throws IOException {
    System.setProperty("org.jline.terminal.dumb", "true");
    List<Path> paths = new ArrayList<>();
    if (input.toFile().isFile()) {
        paths.add(input);
    } else {
        Set<String> dirNamesToProcess = new HashSet<>();
        if (dirList != null) {
            List<String> dirNames = TextIO.loadLines(dirList, "#");
            Log.info("Directory names to process:");
            for (String dirName : dirNames) {
                Log.info(dirName);
            }
            dirNamesToProcess.addAll(dirNames);
        }
        List<Path> directories = Files.walk(input, recurse ? Integer.MAX_VALUE : 1).filter(s -> s.toFile().isDirectory() && !s.equals(input)).collect(Collectors.toList());
        for (Path directory : directories) {
            if (dirList != null && !dirNamesToProcess.contains(directory.toFile().getName())) {
                continue;
            }
            paths.addAll(Files.walk(directory, 1).filter(s -> s.toFile().isFile()).collect(Collectors.toList()));
        }
    }
    Log.info("There are %d files to process.", paths.size());
    long totalLines = 0;
    for (Path path : paths) {
        totalLines += TextIO.lineCount(path);
    }
    if (paths.size() == 0) {
        Log.info("No corpus files found for input : %s", input);
        System.exit(0);
    }
    AtomicLong sentenceCount = new AtomicLong(0);
    try (PrintWriter pw = new PrintWriter(output.toFile(), "UTF-8")) {
        BlockTextLoader loader = BlockTextLoader.fromPaths(paths, 30_000);
        BlockingExecutor executor = new BlockingExecutor(threadCount);
        for (TextChunk chunk : loader) {
            executor.submit(() -> {
                List<String> data = new ArrayList<>(new LinkedHashSet<>(chunk.getData()));
                List<String> sentences = TextCleaner.cleanAndExtractSentences(data);
                sentences = sentences.stream().filter(this::unambiguous).map(s -> toLowercase ? s.toLowerCase(Turkish.LOCALE) : s).collect(Collectors.toList());
                synchronized (this) {
                    sentences.forEach(pw::println);
                    sentenceCount.addAndGet(sentences.size());
                    System.out.println(chunk.size());
                }
            });
        }
        executor.shutdown();
    }
    Log.info("%d sentences are written in %s", sentenceCount.get(), output);
}

Example 3 with BlockingExecutor

use of zemberek.core.concurrency.BlockingExecutor in project zemberek-nlp by ahmetaa.

the class PreprocessTurkishCorpus method run.

@Override
public void run() throws IOException, InterruptedException {
    List<Path> paths = new ArrayList<>();
    if (input.toFile().isFile()) {
        paths.add(input);
    } else {
        Set<String> dirNamesToProcess = new HashSet<>();
        if (dirList != null) {
            List<String> dirNames = TextIO.loadLines(dirList);
            Log.info("Directory names to process:");
            for (String dirName : dirNames) {
                Log.info(dirName);
            }
            dirNamesToProcess.addAll(dirNames);
        }
        List<Path> directories = Files.walk(input, recurse ? Integer.MAX_VALUE : 1).filter(s -> s.toFile().isDirectory() && !s.equals(input)).collect(Collectors.toList());
        for (Path directory : directories) {
            if (dirList != null && !dirNamesToProcess.contains(directory.toFile().getName())) {
                continue;
            }
            paths.addAll(Files.walk(directory, 1).filter(s -> s.toFile().isFile() && (extension == null || s.endsWith(extension))).collect(Collectors.toList()));
        }
    }
    Log.info("There are %d files to process.", paths.size());
    long totalLines = 0;
    for (Path path : paths) {
        totalLines += TextIO.lineCount(path);
    }
    final long total = totalLines;
    if (paths.size() == 0) {
        Log.info("No corpus files found for input : %s", input);
        System.exit(0);
    }
    AtomicLong sentenceCount = new AtomicLong(0);
    if (operation == Operation.LEMMA) {
        morphology = TurkishMorphology.createWithDefaults();
    }
    try (PrintWriter pw = new PrintWriter(output.toFile(), "UTF-8")) {
        BlockTextLoader loader = BlockTextLoader.fromPaths(paths, 10_000);
        BlockingExecutor executor = new BlockingExecutor(threadCount);
        AtomicInteger count = new AtomicInteger(0);
        for (TextChunk chunk : loader) {
            executor.submit(() -> {
                List<String> processed = chunk.getData().stream().filter(// ignore meta tag lines.
                s -> !s.startsWith("<")).map(TextUtil::normalizeSpacesAndSoftHyphens).collect(Collectors.toList());
                List<String> sentences = TurkishSentenceExtractor.DEFAULT.fromParagraphs(processed);
                sentences = sentences.stream().filter(s -> !TextUtil.containsCombiningDiacritics(s)).map(s -> {
                    if (operation == Operation.LEMMA) {
                        return replaceWordsWithLemma(s);
                    } else {
                        return String.join(" ", TurkishTokenizer.DEFAULT.tokenizeToStrings(s));
                    }
                }).map(s -> toLowercase ? s.toLowerCase(Turkish.LOCALE) : s).collect(Collectors.toList());
                synchronized (this) {
                    sentences.forEach(pw::println);
                    sentenceCount.addAndGet(sentences.size());
                    int c = count.addAndGet(chunk.size());
                    System.out.println(String.format("(%d of %d lines) processed.", c, total));
                }
            });
        }
        executor.shutdown();
        executor.awaitTermination(1, TimeUnit.DAYS);
    }
    Log.info("%d sentences are written in %s", sentenceCount.get(), output);
}

Example 4 with BlockingExecutor

use of zemberek.core.concurrency.BlockingExecutor in project zemberek-nlp by ahmetaa.

the class NormalizationScripts method cleanTwitterData.

static void cleanTwitterData(Path in, Path out) throws Exception {
    AnalysisCache cache = AnalysisCache.builder().dynamicCacheSize(300_000, 500_000).build();
    TurkishMorphology morphology = TurkishMorphology.builder().setCache(cache).setLexicon(RootLexicon.getDefault()).disableUnidentifiedTokenAnalyzer().build();
    int threadCount = Runtime.getRuntime().availableProcessors() / 2;
    if (threadCount > 20) {
        threadCount = 20;
    }
    ExecutorService executorService = new BlockingExecutor(threadCount);
    CompletionService<TwitterSaver> service = new ExecutorCompletionService<>(executorService);
    int blockSize = 20_000;
    BlockTextLoader loader = BlockTextLoader.fromPath(in, blockSize);
    Path foreign = Paths.get(out.toString() + ".foreign");
    TwitterSaver saver = new TwitterSaver(out, foreign, blockSize);
    int bc = 0;
    for (TextChunk block : loader) {
        service.submit(new TwitterTask(morphology, saver, block, bc));
        bc++;
    }
    executorService.shutdown();
    executorService.awaitTermination(1, TimeUnit.DAYS);
}

Example 5 with BlockingExecutor

use of zemberek.core.concurrency.BlockingExecutor in project zemberek-nlp by ahmetaa.

the class NormalizationVocabularyGenerator method collectVocabularyHistogram.

Vocabulary collectVocabularyHistogram(BlockTextLoader corpora, int threadCount) throws Exception {
    ExecutorService executorService = new BlockingExecutor(threadCount);
    Vocabulary result = new Vocabulary();
    for (TextChunk chunk : corpora) {
        Log.info("Processing %s", chunk);
        executorService.submit(new WordCollectorTask(chunk, result));
    }
    executorService.shutdown();
    executorService.awaitTermination(1, TimeUnit.DAYS);
    return result;
}