Examples with BoundedBlockPriorityQueue edu.neu.ccs.pyramid.util.BoundedBlockPriorityQueue used on opensource projects

Example 1 with BoundedBlockPriorityQueue

use of edu.neu.ccs.pyramid.util.BoundedBlockPriorityQueue in project pyramid by cheng-li.

the class App1 method ngramSelection.

private static void ngramSelection(Config config, MultiLabelIndex index, String docFilter, Logger logger) throws Exception {
    logger.info("start ngram selection");
    File metaDataFolder = new File(config.getString("output.folder"), "meta_data");
    FeatureLoader.MatchScoreType matchScoreType;
    String matchScoreTypeString = config.getString("train.feature.ngram.matchScoreType");
    String[] indexIds = getDocsForSplitFromQuery(index, config.getString("train.splitQuery"));
    IdTranslator idTranslator = loadIdTranslator(indexIds);
    LabelTranslator labelTranslator = (LabelTranslator) Serialization.deserialize(new File(metaDataFolder, "label_translator.ser"));
    switch(matchScoreTypeString) {
        case "es_original":
            matchScoreType = FeatureLoader.MatchScoreType.ES_ORIGINAL;
            break;
        case "binary":
            matchScoreType = FeatureLoader.MatchScoreType.BINARY;
            break;
        case "frequency":
            matchScoreType = FeatureLoader.MatchScoreType.FREQUENCY;
            break;
        case "tfifl":
            matchScoreType = FeatureLoader.MatchScoreType.TFIFL;
            break;
        default:
            throw new IllegalArgumentException("unknown ngramMatchScoreType");
    }
    double[][] labels = loadLabels(config, index, idTranslator, labelTranslator);
    int numLabels = labels.length;
    int toKeep = config.getInt("train.feature.ngram.selectPerLabel");
    List<BoundedBlockPriorityQueue<Pair<Ngram, Double>>> queues = new ArrayList<>();
    Comparator<Pair<Ngram, Double>> comparator = Comparator.comparing(p -> p.getSecond());
    for (int l = 0; l < numLabels; l++) {
        queues.add(new BoundedBlockPriorityQueue<>(toKeep, comparator));
    }
    FeatureList featureList = (FeatureList) Serialization.deserialize(new File(metaDataFolder, "feature_list.ser"));
    featureList.getAll().stream().parallel().filter(feature -> feature instanceof Ngram).map(feature -> (Ngram) feature).filter(ngram -> ngram.getN() > 1).forEach(ngram -> {
        double[] scores = StumpSelector.scores(index, labels, ngram, idTranslator, matchScoreType, docFilter);
        for (int l = 0; l < numLabels; l++) {
            queues.get(l).add(new Pair<>(ngram, scores[l]));
        }
    });
    Set<Ngram> kept = new HashSet<>();
    StringBuilder stringBuilder = new StringBuilder();
    for (int l = 0; l < numLabels; l++) {
        stringBuilder.append("-------------------------").append("\n");
        stringBuilder.append(labelTranslator.toExtLabel(l)).append(":").append("\n");
        BoundedBlockPriorityQueue<Pair<Ngram, Double>> queue = queues.get(l);
        while (queue.size() > 0) {
            Ngram ngram = queue.poll().getFirst();
            kept.add(ngram);
            stringBuilder.append(ngram.getNgram()).append(", ");
        }
        stringBuilder.append("\n");
    }
    File selectionFile = new File(metaDataFolder, "selected_ngrams.txt");
    FileUtils.writeStringToFile(selectionFile, stringBuilder.toString());
    logger.info("finish ngram selection");
    logger.info("selected ngrams are written to " + selectionFile.getAbsolutePath());
    // after feature selection, overwrite the feature_list.ser file; rename old files
    FeatureList selectedFeatures = new FeatureList();
    for (Feature feature : featureList.getAll()) {
        if (!(feature instanceof Ngram)) {
            selectedFeatures.add(feature);
        }
        if ((feature instanceof Ngram) && ((Ngram) feature).getN() == 1) {
            selectedFeatures.add(feature);
        }
        if ((feature instanceof Ngram) && ((Ngram) feature).getN() > 1 && kept.contains(feature)) {
            selectedFeatures.add(feature);
        }
    }
    FileUtils.copyFile(new File(metaDataFolder, "feature_list.ser"), new File(metaDataFolder, "feature_list_all.ser"));
    FileUtils.copyFile(new File(metaDataFolder, "feature_list.txt"), new File(metaDataFolder, "feature_list_all.txt"));
    Serialization.serialize(selectedFeatures, new File(metaDataFolder, "feature_list.ser"));
    try (BufferedWriter bufferedWriter = new BufferedWriter(new FileWriter(new File(metaDataFolder, "feature_list.txt")))) {
        for (Feature feature : selectedFeatures.getAll()) {
            bufferedWriter.write(feature.toString());
            bufferedWriter.newLine();
        }
    }
}