2 * Licensed to the Apache Software Foundation (ASF) under one or more
3 * contributor license agreements. See the NOTICE file distributed with
4 * this work for additional information regarding copyright ownership.
5 * The ASF licenses this file to You under the Apache License, Version 2.0
6 * (the "License"); you may not use this file except in compliance with
7 * the License. You may obtain a copy of the License at
9 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
18 package org.apache.lucene.misc;
20 import org.apache.lucene.search.DefaultSimilarity;
21 import org.apache.lucene.index.FieldInvertState;
24 import java.util.HashMap;
27 * A similarity with a lengthNorm that provides for a "plateau" of
28 * equally good lengths, and tf helper functions.
31 * For lengthNorm, A global min/max can be specified to define the
32 * plateau of lengths that should all have a norm of 1.0.
33 * Below the min, and above the max the lengthNorm drops off in a
37 * A per field min/max can be specified if different fields have
38 * different sweet spots.
42 * For tf, baselineTf and hyperbolicTf functions are provided, which
43 * subclasses can choose between.
47 public class SweetSpotSimilarity extends DefaultSimilarity {
49 private int ln_min = 1;
50 private int ln_max = 1;
51 private float ln_steep = 0.5f;
53 private Map<String,Number> ln_maxs = new HashMap<String,Number>(7);
54 private Map<String,Number> ln_mins = new HashMap<String,Number>(7);
55 private Map<String,Float> ln_steeps = new HashMap<String,Float>(7);
56 private Map<String,Boolean> ln_overlaps = new HashMap<String,Boolean>(7);
58 private float tf_base = 0.0f;
59 private float tf_min = 0.0f;
61 private float tf_hyper_min = 0.0f;
62 private float tf_hyper_max = 2.0f;
63 private double tf_hyper_base = 1.3d;
64 private float tf_hyper_xoffset = 10.0f;
66 public SweetSpotSimilarity() {
71 * Sets the baseline and minimum function variables for baselineTf
75 public void setBaselineTfFactors(float base, float min) {
81 * Sets the function variables for the hyperbolicTf functions
83 * @param min the minimum tf value to ever be returned (default: 0.0)
84 * @param max the maximum tf value to ever be returned (default: 2.0)
85 * @param base the base value to be used in the exponential for the hyperbolic function (default: e)
86 * @param xoffset the midpoint of the hyperbolic function (default: 10.0)
89 public void setHyperbolicTfFactors(float min, float max,
90 double base, float xoffset) {
94 tf_hyper_xoffset = xoffset;
98 * Sets the default function variables used by lengthNorm when no field
99 * specific variables have been set.
103 public void setLengthNormFactors(int min, int max, float steepness) {
106 this.ln_steep = steepness;
110 * Sets the function variables used by lengthNorm for a specific named field.
112 * @param field field name
113 * @param min minimum value
114 * @param max maximum value
115 * @param steepness steepness of the curve
116 * @param discountOverlaps if true, <code>numOverlapTokens</code> will be
117 * subtracted from <code>numTokens</code>; if false then
118 * <code>numOverlapTokens</code> will be assumed to be 0 (see
119 * {@link DefaultSimilarity#computeNorm(String, FieldInvertState)} for details).
123 public void setLengthNormFactors(String field, int min, int max,
124 float steepness, boolean discountOverlaps) {
125 ln_mins.put(field, Integer.valueOf(min));
126 ln_maxs.put(field, Integer.valueOf(max));
127 ln_steeps.put(field, Float.valueOf(steepness));
128 ln_overlaps.put(field, new Boolean(discountOverlaps));
132 * Implemented as <code> state.getBoost() *
133 * lengthNorm(fieldName, numTokens) </code> where
134 * numTokens does not count overlap tokens if
135 * discountOverlaps is true by default or true for this
138 public float computeNorm(String fieldName, FieldInvertState state) {
140 boolean overlaps = discountOverlaps;
141 if (ln_overlaps.containsKey(fieldName)) {
142 overlaps = ln_overlaps.get(fieldName).booleanValue();
145 numTokens = state.getLength() - state.getNumOverlap();
147 numTokens = state.getLength();
149 return state.getBoost() * computeLengthNorm(fieldName, numTokens);
155 * 1/sqrt( steepness * (abs(x-min) + abs(x-max) - (max-min)) + 1 )
159 * This degrades to <code>1/sqrt(x)</code> when min and max are both 1 and
164 * :TODO: potential optimization is to just flat out return 1.0f if numTerms
165 * is between min and max.
168 * @see #setLengthNormFactors
170 public float computeLengthNorm(String fieldName, int numTerms) {
175 if (ln_mins.containsKey(fieldName)) {
176 l = ln_mins.get(fieldName).intValue();
178 if (ln_maxs.containsKey(fieldName)) {
179 h = ln_maxs.get(fieldName).intValue();
181 if (ln_steeps.containsKey(fieldName)) {
182 s = ln_steeps.get(fieldName).floatValue();
191 (float)(Math.abs(numTerms - l) + Math.abs(numTerms - h) - (h-l))
199 * Delegates to baselineTf
204 public float tf(int freq) {
205 return baselineTf(freq);
211 * (x <= min) ? base : sqrt(x+(base**2)-min)
213 * ...but with a special case check for 0.
215 * This degrates to <code>sqrt(x)</code> when min and base are both 0
218 * @see #setBaselineTfFactors
220 public float baselineTf(float freq) {
222 if (0.0f == freq) return 0.0f;
224 return (freq <= tf_min)
226 : (float)Math.sqrt(freq + (tf_base * tf_base) - tf_min);
230 * Uses a hyperbolic tangent function that allows for a hard max...
233 * tf(x)=min+(max-min)/2*(((base**(x-xoffset)-base**-(x-xoffset))/(base**(x-xoffset)+base**-(x-xoffset)))+1)
237 * This code is provided as a convenience for subclasses that want
238 * to use a hyperbolic tf function.
241 * @see #setHyperbolicTfFactors
243 public float hyperbolicTf(float freq) {
244 if (0.0f == freq) return 0.0f;
246 final float min = tf_hyper_min;
247 final float max = tf_hyper_max;
248 final double base = tf_hyper_base;
249 final float xoffset = tf_hyper_xoffset;
250 final double x = (double)(freq - xoffset);
252 final float result = min +
257 ( ( Math.pow(base,x) - Math.pow(base,-x) )
258 / ( Math.pow(base,x) + Math.pow(base,-x) )
264 return Float.isNaN(result) ? max : result;