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