Search in sources :

Example 1 with Rm3Reranker

use of io.anserini.rerank.rm3.Rm3Reranker in project Anserini by castorini.

the class SearchTweets method main.

public static void main(String[] args) throws Exception {
    long curTime = System.nanoTime();
    SearchArgs searchArgs = new SearchArgs();
    CmdLineParser parser = new CmdLineParser(searchArgs, ParserProperties.defaults().withUsageWidth(90));
    try {
        parser.parseArgument(args);
    } catch (CmdLineException e) {
        System.err.println(e.getMessage());
        parser.printUsage(System.err);
        System.err.println("Example: SearchTweets" + parser.printExample(OptionHandlerFilter.REQUIRED));
        return;
    }
    LOG.info("Reading index at " + searchArgs.index);
    Directory dir;
    if (searchArgs.inmem) {
        LOG.info("Using MMapDirectory with preload");
        dir = new MMapDirectory(Paths.get(searchArgs.index));
        ((MMapDirectory) dir).setPreload(true);
    } else {
        LOG.info("Using default FSDirectory");
        dir = FSDirectory.open(Paths.get(searchArgs.index));
    }
    IndexReader reader = DirectoryReader.open(dir);
    IndexSearcher searcher = new IndexSearcher(reader);
    if (searchArgs.ql) {
        LOG.info("Using QL scoring model");
        searcher.setSimilarity(new LMDirichletSimilarity(searchArgs.mu));
    } else if (searchArgs.bm25) {
        LOG.info("Using BM25 scoring model");
        searcher.setSimilarity(new BM25Similarity(searchArgs.k1, searchArgs.b));
    } else {
        LOG.error("Error: Must specify scoring model!");
        System.exit(-1);
    }
    RerankerCascade cascade = new RerankerCascade();
    if (searchArgs.rm3) {
        cascade.add(new Rm3Reranker(IndexTweets.ANALYZER, StatusField.TEXT.name, "src/main/resources/io/anserini/rerank/rm3/rm3-stoplist.twitter.txt"));
        cascade.add(new RemoveRetweetsTemporalTiebreakReranker());
    } else {
        cascade.add(new RemoveRetweetsTemporalTiebreakReranker());
    }
    if (!searchArgs.model.isEmpty() && searchArgs.extractors != null) {
        LOG.debug(String.format("Ranklib model used, modeled loaded from %s", searchArgs.model));
        cascade.add(new RankLibReranker(searchArgs.model, StatusField.TEXT.name, searchArgs.extractors));
    }
    FeatureExtractors extractorChain = null;
    if (searchArgs.extractors != null) {
        extractorChain = FeatureExtractors.loadExtractor(searchArgs.extractors);
    }
    if (searchArgs.dumpFeatures) {
        PrintStream out = new PrintStream(searchArgs.featureFile);
        Qrels qrels = new Qrels(searchArgs.qrels);
        cascade.add(new TweetsLtrDataGenerator(out, qrels, extractorChain));
    }
    MicroblogTopicSet topics = MicroblogTopicSet.fromFile(new File(searchArgs.topics));
    PrintStream out = new PrintStream(new FileOutputStream(new File(searchArgs.output)));
    LOG.info("Writing output to " + searchArgs.output);
    LOG.info("Initialized complete! (elapsed time = " + (System.nanoTime() - curTime) / 1000000 + "ms)");
    long totalTime = 0;
    int cnt = 0;
    for (MicroblogTopic topic : topics) {
        long curQueryTime = System.nanoTime();
        Query filter = LongPoint.newRangeQuery(StatusField.ID.name, 0L, topic.getQueryTweetTime());
        Query query = AnalyzerUtils.buildBagOfWordsQuery(StatusField.TEXT.name, IndexTweets.ANALYZER, topic.getQuery());
        BooleanQuery.Builder builder = new BooleanQuery.Builder();
        builder.add(filter, BooleanClause.Occur.FILTER);
        builder.add(query, BooleanClause.Occur.MUST);
        Query q = builder.build();
        TopDocs rs = searcher.search(q, searchArgs.hits);
        List<String> queryTokens = AnalyzerUtils.tokenize(IndexTweets.ANALYZER, topic.getQuery());
        RerankerContext context = new RerankerContext(searcher, query, topic.getId(), topic.getQuery(), queryTokens, StatusField.TEXT.name, filter);
        ScoredDocuments docs = cascade.run(ScoredDocuments.fromTopDocs(rs, searcher), context);
        for (int i = 0; i < docs.documents.length; i++) {
            String qid = topic.getId().replaceFirst("^MB0*", "");
            out.println(String.format("%s Q0 %s %d %f %s", qid, docs.documents[i].getField(StatusField.ID.name).numericValue(), (i + 1), docs.scores[i], searchArgs.runtag));
        }
        long qtime = (System.nanoTime() - curQueryTime) / 1000000;
        LOG.info("Query " + topic.getId() + " (elapsed time = " + qtime + "ms)");
        totalTime += qtime;
        cnt++;
    }
    LOG.info("All queries completed!");
    LOG.info("Total elapsed time = " + totalTime + "ms");
    LOG.info("Average query latency = " + (totalTime / cnt) + "ms");
    reader.close();
    out.close();
}
Also used : RemoveRetweetsTemporalTiebreakReranker(io.anserini.rerank.twitter.RemoveRetweetsTemporalTiebreakReranker) ScoredDocuments(io.anserini.rerank.ScoredDocuments) RerankerCascade(io.anserini.rerank.RerankerCascade) Rm3Reranker(io.anserini.rerank.rm3.Rm3Reranker) RankLibReranker(io.anserini.rerank.RankLibReranker) MMapDirectory(org.apache.lucene.store.MMapDirectory) Directory(org.apache.lucene.store.Directory) FSDirectory(org.apache.lucene.store.FSDirectory) PrintStream(java.io.PrintStream) Qrels(io.anserini.util.Qrels) CmdLineParser(org.kohsuke.args4j.CmdLineParser) MMapDirectory(org.apache.lucene.store.MMapDirectory) LongPoint(org.apache.lucene.document.LongPoint) FeatureExtractors(io.anserini.ltr.feature.FeatureExtractors) TweetsLtrDataGenerator(io.anserini.ltr.TweetsLtrDataGenerator) FileOutputStream(java.io.FileOutputStream) IndexReader(org.apache.lucene.index.IndexReader) BM25Similarity(org.apache.lucene.search.similarities.BM25Similarity) LMDirichletSimilarity(org.apache.lucene.search.similarities.LMDirichletSimilarity) File(java.io.File) CmdLineException(org.kohsuke.args4j.CmdLineException) RerankerContext(io.anserini.rerank.RerankerContext)

Example 2 with Rm3Reranker

use of io.anserini.rerank.rm3.Rm3Reranker in project Anserini by castorini.

the class SearchWebCollection method main.

public static void main(String[] args) throws Exception {
    SearchArgs searchArgs = new SearchArgs();
    CmdLineParser parser = new CmdLineParser(searchArgs, ParserProperties.defaults().withUsageWidth(90));
    try {
        parser.parseArgument(args);
    } catch (CmdLineException e) {
        System.err.println(e.getMessage());
        parser.printUsage(System.err);
        System.err.println("Example: SearchWebCollection" + parser.printExample(OptionHandlerFilter.REQUIRED));
        return;
    }
    LOG.info("Reading index at " + searchArgs.index);
    Directory dir;
    if (searchArgs.inmem) {
        LOG.info("Using MMapDirectory with preload");
        dir = new MMapDirectory(Paths.get(searchArgs.index));
        ((MMapDirectory) dir).setPreload(true);
    } else {
        LOG.info("Using default FSDirectory");
        dir = FSDirectory.open(Paths.get(searchArgs.index));
    }
    Similarity similarity = null;
    if (searchArgs.ql) {
        LOG.info("Using QL scoring model");
        similarity = new LMDirichletSimilarity(searchArgs.mu);
    } else if (searchArgs.bm25) {
        LOG.info("Using BM25 scoring model");
        similarity = new BM25Similarity(searchArgs.k1, searchArgs.b);
    } else {
        LOG.error("Error: Must specify scoring model!");
        System.exit(-1);
    }
    RerankerCascade cascade = new RerankerCascade();
    boolean useQueryParser = false;
    if (searchArgs.rm3) {
        cascade.add(new Rm3Reranker(new EnglishAnalyzer(), FIELD_BODY, "src/main/resources/io/anserini/rerank/rm3/rm3-stoplist.gov2.txt"));
        useQueryParser = true;
    } else {
        cascade.add(new IdentityReranker());
    }
    FeatureExtractors extractors = null;
    if (searchArgs.extractors != null) {
        extractors = FeatureExtractors.loadExtractor(searchArgs.extractors);
    }
    if (searchArgs.dumpFeatures) {
        PrintStream out = new PrintStream(searchArgs.featureFile);
        Qrels qrels = new Qrels(searchArgs.qrels);
        cascade.add(new WebCollectionLtrDataGenerator(out, qrels, extractors));
    }
    Path topicsFile = Paths.get(searchArgs.topics);
    if (!Files.exists(topicsFile) || !Files.isRegularFile(topicsFile) || !Files.isReadable(topicsFile)) {
        throw new IllegalArgumentException("Topics file : " + topicsFile + " does not exist or is not a (readable) file.");
    }
    TopicReader tr = (TopicReader) Class.forName("io.anserini.search.query." + searchArgs.topicReader + "TopicReader").getConstructor(Path.class).newInstance(topicsFile);
    SortedMap<Integer, String> topics = tr.read();
    final long start = System.nanoTime();
    SearchWebCollection searcher = new SearchWebCollection(searchArgs.index);
    searcher.search(topics, searchArgs.output, similarity, searchArgs.hits, cascade, useQueryParser, searchArgs.keepstop);
    searcher.close();
    final long durationMillis = TimeUnit.MILLISECONDS.convert(System.nanoTime() - start, TimeUnit.NANOSECONDS);
    LOG.info("Total " + topics.size() + " topics searched in " + DurationFormatUtils.formatDuration(durationMillis, "HH:mm:ss"));
}
Also used : LMDirichletSimilarity(org.apache.lucene.search.similarities.LMDirichletSimilarity) Similarity(org.apache.lucene.search.similarities.Similarity) BM25Similarity(org.apache.lucene.search.similarities.BM25Similarity) IdentityReranker(io.anserini.rerank.IdentityReranker) RerankerCascade(io.anserini.rerank.RerankerCascade) TopicReader(io.anserini.search.query.TopicReader) Rm3Reranker(io.anserini.rerank.rm3.Rm3Reranker) WebCollectionLtrDataGenerator(io.anserini.ltr.WebCollectionLtrDataGenerator) MMapDirectory(org.apache.lucene.store.MMapDirectory) Directory(org.apache.lucene.store.Directory) FSDirectory(org.apache.lucene.store.FSDirectory) Path(java.nio.file.Path) PrintStream(java.io.PrintStream) Qrels(io.anserini.util.Qrels) CmdLineParser(org.kohsuke.args4j.CmdLineParser) EnglishAnalyzer(org.apache.lucene.analysis.en.EnglishAnalyzer) MMapDirectory(org.apache.lucene.store.MMapDirectory) FeatureExtractors(io.anserini.ltr.feature.FeatureExtractors) BM25Similarity(org.apache.lucene.search.similarities.BM25Similarity) LMDirichletSimilarity(org.apache.lucene.search.similarities.LMDirichletSimilarity) CmdLineException(org.kohsuke.args4j.CmdLineException)

Aggregations

FeatureExtractors (io.anserini.ltr.feature.FeatureExtractors)2 RerankerCascade (io.anserini.rerank.RerankerCascade)2 Rm3Reranker (io.anserini.rerank.rm3.Rm3Reranker)2 Qrels (io.anserini.util.Qrels)2 PrintStream (java.io.PrintStream)2 BM25Similarity (org.apache.lucene.search.similarities.BM25Similarity)2 LMDirichletSimilarity (org.apache.lucene.search.similarities.LMDirichletSimilarity)2 Directory (org.apache.lucene.store.Directory)2 FSDirectory (org.apache.lucene.store.FSDirectory)2 MMapDirectory (org.apache.lucene.store.MMapDirectory)2 CmdLineException (org.kohsuke.args4j.CmdLineException)2 CmdLineParser (org.kohsuke.args4j.CmdLineParser)2 TweetsLtrDataGenerator (io.anserini.ltr.TweetsLtrDataGenerator)1 WebCollectionLtrDataGenerator (io.anserini.ltr.WebCollectionLtrDataGenerator)1 IdentityReranker (io.anserini.rerank.IdentityReranker)1 RankLibReranker (io.anserini.rerank.RankLibReranker)1 RerankerContext (io.anserini.rerank.RerankerContext)1 ScoredDocuments (io.anserini.rerank.ScoredDocuments)1 RemoveRetweetsTemporalTiebreakReranker (io.anserini.rerank.twitter.RemoveRetweetsTemporalTiebreakReranker)1 TopicReader (io.anserini.search.query.TopicReader)1