View Javadoc
1   /*
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
8    *
9    *      https://www.apache.org/licenses/LICENSE-2.0
10   *
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.
16   */
17  package org.apache.commons.text.similarity;
18  
19  import java.util.Arrays;
20  
21  /**
22   * An algorithm for measuring the difference between two character sequences using the <a href="https://en.wikipedia.org/wiki/Levenshtein_distance">Levenshtein
23   * Distance</a>.
24   *
25   * <p>
26   * This is the number of changes needed to change one sequence into another, where each change is a single character modification (deletion, insertion or
27   * substitution).
28   * </p>
29   * <p>
30   * This code has been adapted from Apache Commons Lang 3.3.
31   * </p>
32   *
33   * @since 1.0
34   * @see <a href="https://en.wikipedia.org/wiki/Levenshtein_distance">Levenshtein Distance on Wikipedia</a>
35   * @see <a href="https://xlinux.nist.gov/dads/HTML/Levenshtein.html">Levenshtein Distance on NIST</a>
36   */
37  public class LevenshteinDistance implements EditDistance<Integer> {
38  
39      /**
40       * The singleton instance.
41       */
42      private static final LevenshteinDistance INSTANCE = new LevenshteinDistance();
43  
44      /**
45       * Gets the default instance.
46       *
47       * @return The default instance.
48       */
49      public static LevenshteinDistance getDefaultInstance() {
50          return INSTANCE;
51      }
52  
53      /**
54       * Finds the Levenshtein distance between two CharSequences if it's less than or equal to a given threshold.
55       *
56       * <p>
57       * This implementation follows from Algorithms on Strings, Trees and Sequences by Dan Gusfield and Chas Emerick's implementation of the Levenshtein distance
58       * algorithm from <a href="https://www.merriampark.com/ld.htm">http://www.merriampark.com/ld.htm</a>
59       * </p>
60       *
61       * <pre>
62       * limitedCompare(null, *, *)             = Throws {@link IllegalArgumentException}
63       * limitedCompare(*, null, *)             = Throws {@link IllegalArgumentException}
64       * limitedCompare(*, *, -1)               = Throws {@link IllegalArgumentException}
65       * limitedCompare("","", 0)               = 0
66       * limitedCompare("aaapppp", "", 8)       = 7
67       * limitedCompare("aaapppp", "", 7)       = 7
68       * limitedCompare("aaapppp", "", 6))      = -1
69       * limitedCompare("elephant", "hippo", 7) = 7
70       * limitedCompare("elephant", "hippo", 6) = -1
71       * limitedCompare("hippo", "elephant", 7) = 7
72       * limitedCompare("hippo", "elephant", 6) = -1
73       * </pre>
74       *
75       * @param left      the first SimilarityInput, must not be null.
76       * @param right     the second SimilarityInput, must not be null.
77       * @param threshold the target threshold, must not be negative.
78       * @return result distance, or -1
79       */
80      private static <E> int limitedCompare(SimilarityInput<E> left, SimilarityInput<E> right, final int threshold) { // NOPMD
81          if (left == null || right == null) {
82              throw new IllegalArgumentException("CharSequences must not be null");
83          }
84  
85          /*
86           * This implementation only computes the distance if it's less than or equal to the threshold value, returning -1 if it's greater. The advantage is
87           * performance: unbounded distance is O(nm), but a bound of k allows us to reduce it to O(km) time by only computing a diagonal stripe of width 2k + 1
88           * of the cost table. It is also possible to use this to compute the unbounded Levenshtein distance by starting the threshold at 1 and doubling each
89           * time until the distance is found; this is O(dm), where d is the distance.
90           *
91           * One subtlety comes from needing to ignore entries on the border of our stripe, for example,
92           * p[] = |#|#|#|* d[] = *|#|#|#| We must ignore the entry to the left
93           * of the leftmost member We must ignore the entry above the rightmost member
94           *
95           * Another subtlety comes from our stripe running off the matrix if the strings aren't of the same size. Since string s is always swapped to be the
96           * shorter of the two, the stripe will always run off to the upper right instead of the lower left of the matrix.
97           *
98           * As a concrete example, suppose s is of length 5, t is of length 7, and our threshold is 1. In this case we're going to walk a stripe of length 3. The
99           * matrix would look like so:
100          *
101          * <pre> 1 2 3 4 5 1 |#|#| | | | 2 |#|#|#| | | 3 | |#|#|#| | 4 | | |#|#|#| 5 | | | |#|#| 6 | | | | |#| 7 | | | | | | </pre>
102          *
103          * Note how the stripe leads off the table as there is no possible way to turn a string of length 5 into one of length 7 in edit distance of 1.
104          *
105          * Additionally, this implementation decreases memory usage by using two single-dimensional arrays and swapping them back and forth instead of
106          * allocating an entire n by m matrix. This requires a few minor changes, such as immediately returning when it's detected that the stripe has run off
107          * the matrix and initially filling the arrays with large values so that entries we don't compute are ignored.
108          *
109          * See Algorithms on Strings, Trees and Sequences by Dan Gusfield for some discussion.
110          */
111 
112         int n = left.length(); // length of left
113         int m = right.length(); // length of right
114 
115         // if one string is empty, the edit distance is necessarily the length
116         // of the other
117         if (n == 0) {
118             return m <= threshold ? m : -1;
119         }
120         if (m == 0) {
121             return n <= threshold ? n : -1;
122         }
123 
124         if (n > m) {
125             // swap the two strings to consume less memory
126             final SimilarityInput<E> tmp = left;
127             left = right;
128             right = tmp;
129             n = m;
130             m = right.length();
131         }
132 
133         // the edit distance cannot be less than the length difference
134         if (m - n > threshold) {
135             return -1;
136         }
137 
138         int[] p = new int[n + 1]; // 'previous' cost array, horizontally
139         int[] d = new int[n + 1]; // cost array, horizontally
140         int[] tempD; // placeholder to assist in swapping p and d
141 
142         // fill in starting table values
143         final int boundary = Math.min(n, threshold) + 1;
144         for (int i = 0; i < boundary; i++) {
145             p[i] = i;
146         }
147         // these fills ensure that the value above the rightmost entry of our
148         // stripe will be ignored in following loop iterations
149         Arrays.fill(p, boundary, p.length, Integer.MAX_VALUE);
150         Arrays.fill(d, Integer.MAX_VALUE);
151 
152         // iterates through t
153         for (int j = 1; j <= m; j++) {
154             final E rightJ = right.at(j - 1); // jth character of right
155             d[0] = j;
156 
157             // compute stripe indices, constrain to array size
158             final int min = Math.max(1, j - threshold);
159             final int max = j > Integer.MAX_VALUE - threshold ? n : Math.min(n, j + threshold);
160 
161             // ignore entry left of leftmost
162             if (min > 1) {
163                 d[min - 1] = Integer.MAX_VALUE;
164             }
165 
166             int lowerBound = Integer.MAX_VALUE;
167             // iterates through [min, max] in s
168             for (int i = min; i <= max; i++) {
169                 if (left.at(i - 1).equals(rightJ)) {
170                     // diagonally left and up
171                     d[i] = p[i - 1];
172                 } else {
173                     // 1 + minimum of cell to the left, to the top, diagonally
174                     // left and up
175                     d[i] = 1 + Math.min(Math.min(d[i - 1], p[i]), p[i - 1]);
176                 }
177                 lowerBound = Math.min(lowerBound, d[i]);
178             }
179             // if the lower bound is greater than the threshold, then exit early
180             if (lowerBound > threshold) {
181                 return -1;
182             }
183 
184             // copy current distance counts to 'previous row' distance counts
185             tempD = p;
186             p = d;
187             d = tempD;
188         }
189 
190         // if p[n] is greater than the threshold, there's no guarantee on it
191         // being the correct
192         // distance
193         if (p[n] <= threshold) {
194             return p[n];
195         }
196         return -1;
197     }
198 
199     /**
200      * Finds the Levenshtein distance between two Strings.
201      *
202      * <p>
203      * A higher score indicates a greater distance.
204      * </p>
205      *
206      * <p>
207      * The previous implementation of the Levenshtein distance algorithm was from
208      * <a href="https://web.archive.org/web/20120526085419/http://www.merriampark.com/ldjava.htm">
209      * https://web.archive.org/web/20120526085419/http://www.merriampark.com/ldjava.htm</a>
210      * </p>
211      *
212      * <p>
213      * This implementation only need one single-dimensional arrays of length s.length() + 1
214      * </p>
215      *
216      * <pre>
217      * unlimitedCompare(null, *)             = Throws {@link IllegalArgumentException}
218      * unlimitedCompare(*, null)             = Throws {@link IllegalArgumentException}
219      * unlimitedCompare("","")               = 0
220      * unlimitedCompare("","a")              = 1
221      * unlimitedCompare("aaapppp", "")       = 7
222      * unlimitedCompare("frog", "fog")       = 1
223      * unlimitedCompare("fly", "ant")        = 3
224      * unlimitedCompare("elephant", "hippo") = 7
225      * unlimitedCompare("hippo", "elephant") = 7
226      * unlimitedCompare("hippo", "zzzzzzzz") = 8
227      * unlimitedCompare("hello", "hallo")    = 1
228      * </pre>
229      *
230      * @param left  the first CharSequence, must not be null.
231      * @param right the second CharSequence, must not be null.
232      * @return result distance, or -1.
233      * @throws IllegalArgumentException if either CharSequence input is {@code null}.
234      */
235     private static <E> int unlimitedCompare(SimilarityInput<E> left, SimilarityInput<E> right) {
236         if (left == null || right == null) {
237             throw new IllegalArgumentException("CharSequences must not be null");
238         }
239         /*
240          * This implementation use two variable to record the previous cost counts, So this implementation use less memory than previous impl.
241          */
242         int n = left.length(); // length of left
243         int m = right.length(); // length of right
244 
245         if (n == 0) {
246             return m;
247         }
248         if (m == 0) {
249             return n;
250         }
251         if (n > m) {
252             // swap the input strings to consume less memory
253             final SimilarityInput<E> tmp = left;
254             left = right;
255             right = tmp;
256             n = m;
257             m = right.length();
258         }
259         final int[] p = new int[n + 1];
260         // indexes into strings left and right
261         int i; // iterates through left
262         int j; // iterates through right
263         int upperLeft;
264         int upper;
265         E rightJ; // jth character of right
266         int cost; // cost
267         for (i = 0; i <= n; i++) {
268             p[i] = i;
269         }
270         for (j = 1; j <= m; j++) {
271             upperLeft = p[0];
272             rightJ = right.at(j - 1);
273             p[0] = j;
274 
275             for (i = 1; i <= n; i++) {
276                 upper = p[i];
277                 cost = left.at(i - 1).equals(rightJ) ? 0 : 1;
278                 // minimum of cell to the left+1, to the top+1, diagonally left and up +cost
279                 p[i] = Math.min(Math.min(p[i - 1] + 1, p[i] + 1), upperLeft + cost);
280                 upperLeft = upper;
281             }
282         }
283         return p[n];
284     }
285 
286     /**
287      * Threshold.
288      */
289     private final Integer threshold;
290 
291     /**
292      * Constructs a default instance that uses a version of the algorithm that does not use a threshold parameter.
293      *
294      * @see LevenshteinDistance#getDefaultInstance()
295      * @deprecated Use {@link #getDefaultInstance()}.
296      */
297     @Deprecated
298     public LevenshteinDistance() {
299         this(null);
300     }
301 
302     /**
303      * Constructs a new instance. If the threshold is not null, distance calculations will be limited to a maximum length. If the threshold is null, the
304      * unlimited version of the algorithm will be used.
305      *
306      * @param threshold If this is null then distances calculations will not be limited. This may not be negative.
307      */
308     public LevenshteinDistance(final Integer threshold) {
309         if (threshold != null && threshold < 0) {
310             throw new IllegalArgumentException("Threshold must not be negative");
311         }
312         this.threshold = threshold;
313     }
314 
315     /**
316      * Computes the Levenshtein distance between two Strings.
317      *
318      * <p>
319      * A higher score indicates a greater distance.
320      * </p>
321      *
322      * <p>
323      * The previous implementation of the Levenshtein distance algorithm was from
324      * <a href="https://www.merriampark.com/ld.htm">http://www.merriampark.com/ld.htm</a>
325      * </p>
326      *
327      * <p>
328      * Chas Emerick has written an implementation in Java, which avoids an OutOfMemoryError which can occur when my Java implementation is used with very large
329      * strings.<br>
330      * This implementation of the Levenshtein distance algorithm is from
331      * <a href="https://www.merriampark.com/ldjava.htm">http://www.merriampark.com/ldjava.htm</a>
332      * </p>
333      *
334      * <pre>
335      * distance.apply(null, *)             = Throws {@link IllegalArgumentException}
336      * distance.apply(*, null)             = Throws {@link IllegalArgumentException}
337      * distance.apply("","")               = 0
338      * distance.apply("","a")              = 1
339      * distance.apply("aaapppp", "")       = 7
340      * distance.apply("frog", "fog")       = 1
341      * distance.apply("fly", "ant")        = 3
342      * distance.apply("elephant", "hippo") = 7
343      * distance.apply("hippo", "elephant") = 7
344      * distance.apply("hippo", "zzzzzzzz") = 8
345      * distance.apply("hello", "hallo")    = 1
346      * </pre>
347      *
348      * @param left  the first input, must not be null.
349      * @param right the second input, must not be null.
350      * @return result distance, or -1.
351      * @throws IllegalArgumentException if either String input {@code null}.
352      */
353     @Override
354     public Integer apply(final CharSequence left, final CharSequence right) {
355         return apply(SimilarityInput.input(left), SimilarityInput.input(right));
356     }
357 
358     /**
359      * Computes the Levenshtein distance between two inputs.
360      *
361      * <p>
362      * A higher score indicates a greater distance.
363      * </p>
364      *
365      * <pre>
366      * distance.apply(null, *)             = Throws {@link IllegalArgumentException}
367      * distance.apply(*, null)             = Throws {@link IllegalArgumentException}
368      * distance.apply("","")               = 0
369      * distance.apply("","a")              = 1
370      * distance.apply("aaapppp", "")       = 7
371      * distance.apply("frog", "fog")       = 1
372      * distance.apply("fly", "ant")        = 3
373      * distance.apply("elephant", "hippo") = 7
374      * distance.apply("hippo", "elephant") = 7
375      * distance.apply("hippo", "zzzzzzzz") = 8
376      * distance.apply("hello", "hallo")    = 1
377      * </pre>
378      *
379      * @param <E>   The type of similarity score unit.
380      * @param left  the first input, must not be null.
381      * @param right the second input, must not be null.
382      * @return result distance, or -1.
383      * @throws IllegalArgumentException if either String input {@code null}.
384      * @since 1.13.0
385      */
386     public <E> Integer apply(final SimilarityInput<E> left, final SimilarityInput<E> right) {
387         if (threshold != null) {
388             return limitedCompare(left, right, threshold);
389         }
390         return unlimitedCompare(left, right);
391     }
392 
393     /**
394      * Gets the distance threshold.
395      *
396      * @return The distance threshold.
397      */
398     public Integer getThreshold() {
399         return threshold;
400     }
401 
402 }