Examples with TextUtil zemberek.core.text.TextUtil used on opensource projects

Example 1 with TextUtil

use of zemberek.core.text.TextUtil 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);
}