Lucene CustomScoreQuery 不从 FunctionQuery 的 FieldSource 传递值

Question

如果我从 Lucene Java Doc Page 中理解正确，将 CustomScoreQuery 实例设置为严格应该传递 FunctionQuery 的 FieldSource 值而不修改（如规范化）到方法 public float customScore(int doc, float subQueryScore, float valSrcScore) 中 CustomScoreProvider 的 valSrcScore。因此，我想，我得到了精确的浮点值，它存储在文档的 FloatSourceField 中。

但当索引数据量变大时，情况似乎并非如此。这里我有一个最小的例子来说明我的意思：

import org.apache.lucene.analysis.standard.StandardAnalyzer;
import org.apache.lucene.document.*;
import org.apache.lucene.index.*;
import org.apache.lucene.queries.*;
import org.apache.lucene.queries.function.FunctionQuery;
import org.apache.lucene.queries.function.valuesource.FloatFieldSource;
import org.apache.lucene.search.*;
import org.apache.lucene.store.RAMDirectory;
import org.apache.lucene.util.Version;
import java.io.IOException;
public class CustomScoreTest {
    public static void main(String[] args) throws IOException {
        RAMDirectory index = new RAMDirectory();
        IndexWriterConfig config = new IndexWriterConfig(Version.LATEST, new StandardAnalyzer());
        IndexWriter writer = new IndexWriter(index, config);

        // prepare dummy text
        String text = "";
        for (int i = 0; i < 1000; i++) text += "abc ";

        // add dummy docs
        for (int i = 0; i <25000; i++) {
            Document doc = new Document();
            doc.add(new FloatField("number", i * 100f, Field.Store.YES));
            doc.add(new TextField("text", text, Field.Store.YES));
            writer.addDocument(doc);
        }
        writer.close();

        IndexReader reader = IndexReader.open(index);
        IndexSearcher searcher = new IndexSearcher(reader);

        Query q1 = new TermQuery(new Term("text", "abc"));
        CustomScoreQuery q2 = new CustomScoreQuery(q1, new FunctionQuery(new FloatFieldSource("number"))) {
            protected CustomScoreProvider getCustomScoreProvider(AtomicReaderContext ctx) throws IOException {
                return new CustomScoreProvider(ctx) {
                    public float customScore(int doc, float subQueryScore, float valSrcScore) throws IOException {
                        float diff = Math.abs(valSrcScore - searcher.doc(doc).getField("number").numericValue().floatValue());
                        if (diff > 0) throw new IllegalStateException("diff: " + diff);
                        return super.customScore(doc, subQueryScore, valSrcScore);
                    }
                };
            }
        };

        // In strict custom scoring, the part does not participate in weight normalization.
        // This may be useful when one wants full control over how scores are modified, and
        // does not care about normalising by the  part
        q2.setStrict(true);

        // Exception in thread "main" java.lang.IllegalStateException: diff: 1490700.0
        searcher.search(q2, 10);
    }
}

如本示例中所述，抛出异常是因为 valSrcScore 与存储在文档 "number" 字段中的实际值有很大差异。

但是当我将索引虚拟文档的数量减少到 2500 个时，它按预期工作并且我得到的值与 "number" 字段中的值相差 0。

我在这里做错了什么？

Answer 1

你是哪个版本的 lucene 运行？一种可能是随着索引大小的增长，AtomicReaderContext 应该替换为 LeafReaderContext。只是假设

Lucene CustomScoreQuery 不从 FunctionQuery 的 FieldSource 传递值

Lucene CustomScoreQuery does not pass value from FunctionQuery's FieldSource

java

lucene