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

Example 6 with TextChunk

use of zemberek.core.text.TextChunk in project zemberek-nlp by ahmetaa.

the class ItemFindExperiment method processCorpus.

static void processCorpus(Path in, Path out) throws IOException {
    BlockTextLoader loader = BlockTextLoader.fromPath(in, 10000);
    try (PrintWriter pw = new PrintWriter(out.toFile(), "utf-8")) {
        for (TextChunk chunk : loader) {
            LinkedHashSet<String> unique = new LinkedHashSet<>(chunk.getData());
            for (String l : unique) {
                if (!Strings.containsNone(l, "[]#~|")) {
                    continue;
                }
                l = l.toLowerCase(es).replaceAll("[^0-9a-zñáéíóúü]", " ").replaceAll("\\s+", " ").trim();
                if (l.length() == 0) {
                    continue;
                }
                if (l.length() < 20) {
                    continue;
                }
                pw.println(l);
            }
        }
    }
}

Example 7 with TextChunk

use of zemberek.core.text.TextChunk 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 8 with TextChunk

use of zemberek.core.text.TextChunk 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 9 with TextChunk

use of zemberek.core.text.TextChunk in project zemberek-nlp by ahmetaa.

the class Dictionary method readFromFile.

static Dictionary readFromFile(Path file, final Args args) {
    Log.info("Initialize dictionary and histograms.");
    Dictionary dictionary = new Dictionary(args);
    Log.info("Loading text.");
    BlockTextLoader loader = BlockTextLoader.fromPath(file, 100_000);
    SpaceTabTokenizer tokenizer = new SpaceTabTokenizer();
    int blockCounter = 1;
    for (TextChunk lines : loader) {
        for (String line : lines) {
            List<String> split = tokenizer.splitToList(line);
            split.add(EOS);
            for (String word : split) {
                if (word.startsWith("#")) {
                    continue;
                }
                dictionary.add(word);
            }
        }
        Log.info("Lines read: %d (thousands) ", blockCounter * 100);
        blockCounter++;
    }
    Log.info("Word + Label count = %d", dictionary.words_.size());
    Log.info("Removing word and labels with small counts. Min word = %d, Min Label = %d", args.minCount, args.minCountLabel);
    // now we have the histograms. Remove based on count.
    dictionary.words_.sort((e1, e2) -> {
        if (e1.type != e2.type) {
            return Integer.compare(e1.type, e2.type);
        } else {
            return Long.compare(e2.count, e1.count);
        }
    });
    // TODO: add threshold method.
    LinkedHashSet<Entry> all = new LinkedHashSet<>(dictionary.words_);
    List<Entry> toRemove = dictionary.words_.stream().filter(s -> (s.type == TYPE_WORD && s.count < args.minCount || s.type == TYPE_LABEL && s.count < args.minCountLabel)).collect(Collectors.toList());
    all.removeAll(toRemove);
    dictionary.words_ = new ArrayList<>(all);
    dictionary.size_ = 0;
    dictionary.nwords_ = 0;
    dictionary.nlabels_ = 0;
    Arrays.fill(dictionary.word2int_, -1);
    for (Entry e : dictionary.words_) {
        int i = dictionary.find(e.word);
        dictionary.word2int_[i] = dictionary.size_++;
        if (e.type == TYPE_WORD) {
            dictionary.nwords_++;
        }
        if (e.type == TYPE_LABEL) {
            dictionary.nlabels_++;
        }
    }
    Log.info("Word count = %d , Label count = %d", dictionary.nwords(), dictionary.nlabels());
    dictionary.initTableDiscard();
    dictionary.initNgrams();
    return dictionary;
}

Example 10 with TextChunk

use of zemberek.core.text.TextChunk in project zemberek-nlp by ahmetaa.

the class ProcessTwnertcData method main.

public static void main(String[] args) throws IOException {
    Path corpus = Paths.get("/media/ahmetaa/depo/ner/TWNERTC_All_Versions/TWNERTC_TC_Coarse_Grained_NER_DomainDependent_NoiseReduction.DUMP");
    Path nerOut = Paths.get("/media/ahmetaa/depo/ner/ner-coarse");
    Path categoryOut = Paths.get("/media/ahmetaa/depo/classification/twnertc-data");
    BlockTextLoader loader = BlockTextLoader.fromPath(corpus, 10_000);
    List<String> nerLines = new ArrayList<>();
    List<String> categoryLines = new ArrayList<>();
    Histogram<String> categories = new Histogram<>();
    for (TextChunk chunk : loader) {
        for (String line : chunk) {
            List<String> parts = TextUtil.TAB_SPLITTER.splitToList(line);
            categoryLines.add("__label__" + parts.get(0) + " " + parts.get(2));
            categories.add(parts.get(0));
            List<String> nerLabels = TextUtil.SPACE_SPLITTER.splitToList(parts.get(1));
            List<String> nerWords = TextUtil.SPACE_SPLITTER.splitToList(parts.get(2));
            if (nerLabels.size() != nerWords.size()) {
                continue;
            }
            List<NerRange> ranges = new ArrayList<>();
            NerRange range = new NerRange();
            for (int i = 0; i < nerLabels.size(); i++) {
                String lbl = nerLabels.get(i);
                String word = nerWords.get(i);
                if (lbl.equals("O")) {
                    if (range.type == null) {
                        range.type = "O";
                    } else {
                        if (range.type.equals("O")) {
                            range.seq.add(word);
                        } else {
                            ranges.add(range);
                            range = new NerRange();
                            range.type = "O";
                            range.seq.add(word);
                        }
                    }
                }
            }
        }
        Log.info(chunk.index * loader.getBlockSize());
    }
    Files.write(categoryOut, categoryLines);
    categories.saveSortedByCounts(Paths.get("/media/ahmetaa/depo/classification/categories"), " ");
}