feat(compression): Add Burrows-Wheeler Transform (BWT) and Move-to-Front (MTF) (#6926)

* feat(compression): Add Burrows-Wheeler Transform (BWT) and Move-to-Front (MTF)

* Resolve SpotBugs

* fix code style
This commit is contained in:
Indolyn Yi
2025-10-25 17:29:45 +08:00
committed by GitHub
parent 48ba1ae466
commit ab65ac6485
4 changed files with 600 additions and 0 deletions

View File

@@ -0,0 +1,220 @@
package com.thealgorithms.compression;
import java.util.Arrays;
import java.util.HashMap;
import java.util.Map;
/**
* Implementation of the Burrows-Wheeler Transform (BWT) and its inverse.
* <p>
* BWT is a reversible data transformation algorithm that rearranges a string into runs of
* similar characters. While not a compression algorithm itself, it significantly improves
* the compressibility of data for subsequent algorithms like Move-to-Front encoding and
* Run-Length Encoding.
* </p>
*
* <p>The transform works by:
* <ol>
* <li>Generating all rotations of the input string</li>
* <li>Sorting these rotations lexicographically</li>
* <li>Taking the last column of the sorted matrix as output</li>
* <li>Recording the index of the original string in the sorted matrix</li>
* </ol>
* </p>
*
* <p><b>Important:</b> The input string should end with a unique end-of-string marker
* (typically '$') that:
* <ul>
* <li>Does not appear anywhere else in the text</li>
* <li>Is lexicographically smaller than all other characters</li>
* <li>Ensures unique rotations and enables correct inverse transformation</li>
* </ul>
* Without this marker, the inverse transform may not correctly reconstruct the original string.
* </p>
*
* <p><b>Time Complexity:</b>
* <ul>
* <li>Forward transform: O(n² log n) where n is the string length</li>
* <li>Inverse transform: O(n) using the LF-mapping technique</li>
* </ul>
* </p>
*
* <p><b>Example:</b></p>
* <pre>
* Input: "banana$"
* Output: BWTResult("annb$aa", 4)
* - "annb$aa" is the transformed string (groups similar characters)
* - 4 is the index of the original string in the sorted rotations
* </pre>
*
* @see <a href="https://en.wikipedia.org/wiki/Burrows%E2%80%93Wheeler_transform">BurrowsWheeler transform (Wikipedia)</a>
*/
public final class BurrowsWheelerTransform {
private BurrowsWheelerTransform() {
}
/**
* A container for the result of the forward BWT.
* <p>
* Contains the transformed string and the index of the original string
* in the sorted rotations matrix, both of which are required for the
* inverse transformation.
* </p>
*/
public static class BWTResult {
/** The transformed string (last column of the sorted rotation matrix) */
public final String transformed;
/** The index of the original string in the sorted rotations matrix */
public final int originalIndex;
/**
* Constructs a BWTResult with the transformed string and original index.
*
* @param transformed the transformed string (L-column)
* @param originalIndex the index of the original string in sorted rotations
*/
public BWTResult(String transformed, int originalIndex) {
this.transformed = transformed;
this.originalIndex = originalIndex;
}
@Override
public boolean equals(Object obj) {
if (this == obj) {
return true;
}
if (obj == null || getClass() != obj.getClass()) {
return false;
}
BWTResult bwtResult = (BWTResult) obj;
return originalIndex == bwtResult.originalIndex && transformed.equals(bwtResult.transformed);
}
@Override
public int hashCode() {
return 31 * transformed.hashCode() + originalIndex;
}
@Override
public String toString() {
return "BWTResult[transformed=" + transformed + ", originalIndex=" + originalIndex + "]";
}
}
/**
* Performs the forward Burrows-Wheeler Transform on the input string.
* <p>
* The algorithm generates all cyclic rotations of the input, sorts them
* lexicographically, and returns the last column of this sorted matrix
* along with the position of the original string.
* </p>
*
* <p><b>Note:</b> It is strongly recommended that the input string ends with
* a unique end-of-string marker (e.g., '$') that is lexicographically smaller
* than any other character in the string. This ensures correct inversion.</p>
*
* @param text the input string to transform; must not be {@code null}
* @return a {@link BWTResult} object containing the transformed string (L-column)
* and the index of the original string in the sorted rotations matrix;
* returns {@code BWTResult("", -1)} for empty input
* @throws NullPointerException if {@code text} is {@code null}
*/
public static BWTResult transform(String text) {
if (text == null || text.isEmpty()) {
return new BWTResult("", -1);
}
int n = text.length();
// Generate all rotations of the input string
String[] rotations = new String[n];
for (int i = 0; i < n; i++) {
rotations[i] = text.substring(i) + text.substring(0, i);
}
// Sort rotations lexicographically
Arrays.sort(rotations);
int originalIndex = Arrays.binarySearch(rotations, text);
StringBuilder lastColumn = new StringBuilder(n);
for (int i = 0; i < n; i++) {
lastColumn.append(rotations[i].charAt(n - 1));
}
return new BWTResult(lastColumn.toString(), originalIndex);
}
/**
* Performs the inverse Burrows-Wheeler Transform using the LF-mapping technique.
* <p>
* The LF-mapping (Last-First mapping) is an efficient method to reconstruct
* the original string from the BWT output without explicitly reconstructing
* the entire sorted rotations matrix.
* </p>
*
* <p>The algorithm works by:
* <ol>
* <li>Creating the first column by sorting the BWT string</li>
* <li>Building a mapping from first column indices to last column indices</li>
* <li>Following this mapping starting from the original index to reconstruct the string</li>
* </ol>
* </p>
*
* @param bwtString the transformed string (L-column) from the forward transform; must not be {@code null}
* @param originalIndex the index of the original string row from the forward transform;
* use -1 for empty strings
* @return the original, untransformed string; returns empty string if input is empty or {@code originalIndex} is -1
* @throws NullPointerException if {@code bwtString} is {@code null}
* @throws IllegalArgumentException if {@code originalIndex} is out of valid range (except -1)
*/
public static String inverseTransform(String bwtString, int originalIndex) {
if (bwtString == null || bwtString.isEmpty() || originalIndex == -1) {
return "";
}
int n = bwtString.length();
if (originalIndex < 0 || originalIndex >= n) {
throw new IllegalArgumentException("Original index must be between 0 and " + (n - 1) + ", got: " + originalIndex);
}
char[] lastColumn = bwtString.toCharArray();
char[] firstColumn = bwtString.toCharArray();
Arrays.sort(firstColumn);
// Create the "next" array for LF-mapping.
// next[i] stores the row index in the last column that corresponds to firstColumn[i]
int[] next = new int[n];
// Track the count of each character seen so far in the last column
Map<Character, Integer> countMap = new HashMap<>();
// Store the first occurrence index of each character in the first column
Map<Character, Integer> firstOccurrence = new HashMap<>();
for (int i = 0; i < n; i++) {
if (!firstOccurrence.containsKey(firstColumn[i])) {
firstOccurrence.put(firstColumn[i], i);
}
}
// Build the LF-mapping
for (int i = 0; i < n; i++) {
char c = lastColumn[i];
int count = countMap.getOrDefault(c, 0);
int firstIndex = firstOccurrence.get(c);
next[firstIndex + count] = i;
countMap.put(c, count + 1);
}
// Reconstruct the original string by following the LF-mapping
StringBuilder originalString = new StringBuilder(n);
int currentRow = originalIndex;
for (int i = 0; i < n; i++) {
originalString.append(firstColumn[currentRow]);
currentRow = next[currentRow];
}
return originalString.toString();
}
}

View File

@@ -0,0 +1,164 @@
package com.thealgorithms.compression;
import java.util.ArrayList;
import java.util.Collection;
import java.util.LinkedList;
import java.util.List;
import java.util.stream.Collectors;
/**
* Implementation of the Move-to-Front (MTF) transform and its inverse.
* <p>
* MTF is a data transformation algorithm that encodes each symbol in the input
* as its current position in a dynamically-maintained list, then moves that symbol
* to the front of the list. This transformation is particularly effective when used
* after the Burrows-Wheeler Transform (BWT), as BWT groups similar characters together.
* </p>
*
* <p>The transform converts runs of repeated characters into sequences of small integers
* (often zeros), which are highly compressible by subsequent entropy encoding algorithms
* like Run-Length Encoding (RLE) or Huffman coding. This technique is used in the
* bzip2 compression algorithm.
* </p>
*
* <p><b>How it works:</b>
* <ol>
* <li>Maintain a list of symbols (the alphabet), initially in a fixed order</li>
* <li>For each input symbol:
* <ul>
* <li>Output its current index in the list</li>
* <li>Move that symbol to the front of the list</li>
* </ul>
* </li>
* </ol>
* This means frequently occurring symbols quickly move to the front and are encoded
* with small indices (often 0), while rare symbols remain near the back.
* </p>
*
* <p><b>Time Complexity:</b>
* <ul>
* <li>Forward transform: O(n × m) where n is input length and m is alphabet size</li>
* <li>Inverse transform: O(n × m)</li>
* </ul>
* Note: Using {@link LinkedList} for O(1) insertions and O(m) search operations.
* </p>
*
* <p><b>Example:</b></p>
* <pre>
* Input: "annb$aa"
* Alphabet: "$abn" (initial order)
* Output: [1, 3, 0, 3, 3, 3, 0]
*
* Step-by-step:
* - 'a': index 1 in [$,a,b,n] → output 1, list becomes [a,$,b,n]
* - 'n': index 3 in [a,$,b,n] → output 3, list becomes [n,a,$,b]
* - 'n': index 0 in [n,a,$,b] → output 0, list stays [n,a,$,b]
* - 'b': index 3 in [n,a,$,b] → output 3, list becomes [b,n,a,$]
* - etc.
*
* Notice how repeated 'n' characters produce zeros after the first occurrence!
* </pre>
*
* @see <a href="https://en.wikipedia.org/wiki/Move-to-front_transform">Move-to-front transform (Wikipedia)</a>
*/
public final class MoveToFront {
private MoveToFront() {
}
/**
* Performs the forward Move-to-Front transform.
* <p>
* Converts the input string into a list of integers, where each integer represents
* the position of the corresponding character in a dynamically-maintained alphabet list.
* </p>
*
* <p><b>Note:</b> All characters in the input text must exist in the provided alphabet,
* otherwise an {@link IllegalArgumentException} is thrown. The alphabet should contain
* all unique characters that may appear in the input.</p>
*
* @param text the input string to transform; if empty, returns an empty list
* @param initialAlphabet a string containing the initial ordered set of symbols
* (e.g., "$abn" or the full ASCII set); must not be empty
* when {@code text} is non-empty
* @return a list of integers representing the transformed data, where each integer
* is the index of the corresponding input character in the current alphabet state
* @throws IllegalArgumentException if {@code text} is non-empty and {@code initialAlphabet}
* is {@code null} or empty
* @throws IllegalArgumentException if any character in {@code text} is not found in
* {@code initialAlphabet}
*/
public static List<Integer> transform(String text, String initialAlphabet) {
if (text == null || text.isEmpty()) {
return new ArrayList<>();
}
if (initialAlphabet == null || initialAlphabet.isEmpty()) {
throw new IllegalArgumentException("Alphabet cannot be null or empty when text is not empty.");
}
List<Integer> output = new ArrayList<>(text.length());
// Use LinkedList for O(1) add-to-front and O(n) remove operations
// This is more efficient than ArrayList for the move-to-front pattern
List<Character> alphabet = initialAlphabet.chars().mapToObj(c -> (char) c).collect(Collectors.toCollection(LinkedList::new));
for (char c : text.toCharArray()) {
int index = alphabet.indexOf(c);
if (index == -1) {
throw new IllegalArgumentException("Symbol '" + c + "' not found in the initial alphabet.");
}
output.add(index);
// Move the character to the front
Character symbol = alphabet.remove(index);
alphabet.addFirst(symbol);
}
return output;
}
/**
* Performs the inverse Move-to-Front transform.
* <p>
* Reconstructs the original string from the list of indices produced by the
* forward transform. This requires the exact same initial alphabet that was
* used in the forward transform.
* </p>
*
* <p><b>Important:</b> The {@code initialAlphabet} parameter must be identical
* to the one used in the forward transform, including character order, or the
* output will be incorrect.</p>
*
* @param indices The list of integers from the forward transform.
* @param initialAlphabet the exact same initial alphabet string used for the forward transform;
* if {@code null} or empty, returns an empty string
* @return the original, untransformed string
* @throws IllegalArgumentException if any index in {@code indices} is negative or
* exceeds the current alphabet size
*/
public static String inverseTransform(Collection<Integer> indices, String initialAlphabet) {
if (indices == null || indices.isEmpty() || initialAlphabet == null || initialAlphabet.isEmpty()) {
return "";
}
StringBuilder output = new StringBuilder(indices.size());
// Use LinkedList for O(1) add-to-front and O(n) remove operations
List<Character> alphabet = initialAlphabet.chars().mapToObj(c -> (char) c).collect(Collectors.toCollection(LinkedList::new));
for (int index : indices) {
if (index < 0 || index >= alphabet.size()) {
throw new IllegalArgumentException("Index " + index + " is out of bounds for the current alphabet of size " + alphabet.size() + ".");
}
// Get the symbol at the index
char symbol = alphabet.get(index);
output.append(symbol);
// Move the symbol to the front (mirroring the forward transform)
alphabet.remove(index);
alphabet.addFirst(symbol);
}
return output.toString();
}
}