BM25Stats.java

package org.apache.lucene.search.similarities;

import org.apache.lucene.search.Explanation;

public class BM25Stats {
	/** BM25's idf */
    private final Explanation idf;
    /** The average document length. */
    private final float avgdl;
    /** query's inner boost */
    private final float queryBoost;
    /** query's outer boost (only for explain) */
    private float topLevelBoost;
    /** weight (idf * boost) */
    private float weight;
    /** field name, for pulling norms */
    private final String field;
    /** precomputed norm[256] with k1 * ((1 - b) + b * dl / avgdl) */
    private final float cache[];

    BM25Stats(String field, Explanation idf, float queryBoost, float avgdl, float cache[]) {
      this.field = field;
      this.idf = idf;
      this.queryBoost = queryBoost;
      this.avgdl = avgdl;
      this.cache = cache;
    }

    public float getValueForNormalization() {
      // we return a TF-IDF like normalization to be nice, but we don't actually normalize ourselves.
      final float queryWeight = idf.getValue() * queryBoost;
      return queryWeight * queryWeight;
    }

    public void normalize(float queryNorm, float topLevelBoost) {
      // we don't normalize with queryNorm at all, we just capture the top-level boost
      this.topLevelBoost = topLevelBoost;
      this.weight = idf.getValue() * queryBoost * topLevelBoost;
    } 
}