FusionDictionary.java revision 51a0ef8c59ea590b6e5e80a82fc75bf244084270
1/* 2 * Copyright (C) 2011 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); you may not 5 * use this file except in compliance with the License. You may obtain a copy of 6 * the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT 12 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the 13 * License for the specific language governing permissions and limitations under 14 * the License. 15 */ 16 17package com.android.inputmethod.latin.makedict; 18 19import com.android.inputmethod.latin.Constants; 20 21import java.util.ArrayList; 22import java.util.Arrays; 23import java.util.Collections; 24import java.util.Date; 25import java.util.HashMap; 26import java.util.Iterator; 27import java.util.LinkedList; 28 29/** 30 * A dictionary that can fusion heads and tails of words for more compression. 31 */ 32public final class FusionDictionary implements Iterable<Word> { 33 private static final boolean DBG = MakedictLog.DBG; 34 35 /** 36 * A node of the dictionary, containing several CharGroups. 37 * 38 * A node is but an ordered array of CharGroups, which essentially contain all the 39 * real information. 40 * This class also contains fields to cache size and address, to help with binary 41 * generation. 42 */ 43 public static final class Node { 44 ArrayList<CharGroup> mData; 45 // To help with binary generation 46 int mCachedSize = Integer.MIN_VALUE; 47 int mCachedAddress = Integer.MIN_VALUE; 48 int mCachedParentAddress = 0; 49 50 public Node() { 51 mData = new ArrayList<CharGroup>(); 52 } 53 public Node(ArrayList<CharGroup> data) { 54 mData = data; 55 } 56 } 57 58 /** 59 * A string with a frequency. 60 * 61 * This represents an "attribute", that is either a bigram or a shortcut. 62 */ 63 public static final class WeightedString { 64 public final String mWord; 65 public int mFrequency; 66 public WeightedString(String word, int frequency) { 67 mWord = word; 68 mFrequency = frequency; 69 } 70 71 @Override 72 public int hashCode() { 73 return Arrays.hashCode(new Object[] { mWord, mFrequency }); 74 } 75 76 @Override 77 public boolean equals(Object o) { 78 if (o == this) return true; 79 if (!(o instanceof WeightedString)) return false; 80 WeightedString w = (WeightedString)o; 81 return mWord.equals(w.mWord) && mFrequency == w.mFrequency; 82 } 83 } 84 85 /** 86 * A group of characters, with a frequency, shortcut targets, bigrams, and children. 87 * 88 * This is the central class of the in-memory representation. A CharGroup is what can 89 * be seen as a traditional "trie node", except it can hold several characters at the 90 * same time. A CharGroup essentially represents one or several characters in the middle 91 * of the trie trie; as such, it can be a terminal, and it can have children. 92 * In this in-memory representation, whether the CharGroup is a terminal or not is represented 93 * in the frequency, where NOT_A_TERMINAL (= -1) means this is not a terminal and any other 94 * value is the frequency of this terminal. A terminal may have non-null shortcuts and/or 95 * bigrams, but a non-terminal may not. Moreover, children, if present, are null. 96 */ 97 public static final class CharGroup { 98 public static final int NOT_A_TERMINAL = -1; 99 final int mChars[]; 100 ArrayList<WeightedString> mShortcutTargets; 101 ArrayList<WeightedString> mBigrams; 102 int mFrequency; // NOT_A_TERMINAL == mFrequency indicates this is not a terminal. 103 Node mChildren; 104 boolean mIsNotAWord; // Only a shortcut 105 boolean mIsBlacklistEntry; 106 // The two following members to help with binary generation 107 int mCachedSize; 108 int mCachedAddress; 109 110 public CharGroup(final int[] chars, final ArrayList<WeightedString> shortcutTargets, 111 final ArrayList<WeightedString> bigrams, final int frequency, 112 final boolean isNotAWord, final boolean isBlacklistEntry) { 113 mChars = chars; 114 mFrequency = frequency; 115 mShortcutTargets = shortcutTargets; 116 mBigrams = bigrams; 117 mChildren = null; 118 mIsNotAWord = isNotAWord; 119 mIsBlacklistEntry = isBlacklistEntry; 120 } 121 122 public CharGroup(final int[] chars, final ArrayList<WeightedString> shortcutTargets, 123 final ArrayList<WeightedString> bigrams, final int frequency, 124 final boolean isNotAWord, final boolean isBlacklistEntry, final Node children) { 125 mChars = chars; 126 mFrequency = frequency; 127 mShortcutTargets = shortcutTargets; 128 mBigrams = bigrams; 129 mChildren = children; 130 mIsNotAWord = isNotAWord; 131 mIsBlacklistEntry = isBlacklistEntry; 132 } 133 134 public void addChild(CharGroup n) { 135 if (null == mChildren) { 136 mChildren = new Node(); 137 } 138 mChildren.mData.add(n); 139 } 140 141 public boolean isTerminal() { 142 return NOT_A_TERMINAL != mFrequency; 143 } 144 145 public int getFrequency() { 146 return mFrequency; 147 } 148 149 public boolean getIsNotAWord() { 150 return mIsNotAWord; 151 } 152 153 public boolean getIsBlacklistEntry() { 154 return mIsBlacklistEntry; 155 } 156 157 public ArrayList<WeightedString> getShortcutTargets() { 158 // We don't want write permission to escape outside the package, so we return a copy 159 if (null == mShortcutTargets) return null; 160 final ArrayList<WeightedString> copyOfShortcutTargets = 161 new ArrayList<WeightedString>(mShortcutTargets); 162 return copyOfShortcutTargets; 163 } 164 165 public ArrayList<WeightedString> getBigrams() { 166 // We don't want write permission to escape outside the package, so we return a copy 167 if (null == mBigrams) return null; 168 final ArrayList<WeightedString> copyOfBigrams = new ArrayList<WeightedString>(mBigrams); 169 return copyOfBigrams; 170 } 171 172 public boolean hasSeveralChars() { 173 assert(mChars.length > 0); 174 return 1 < mChars.length; 175 } 176 177 /** 178 * Adds a word to the bigram list. Updates the frequency if the word already 179 * exists. 180 */ 181 public void addBigram(final String word, final int frequency) { 182 if (mBigrams == null) { 183 mBigrams = new ArrayList<WeightedString>(); 184 } 185 WeightedString bigram = getBigram(word); 186 if (bigram != null) { 187 bigram.mFrequency = frequency; 188 } else { 189 bigram = new WeightedString(word, frequency); 190 mBigrams.add(bigram); 191 } 192 } 193 194 /** 195 * Gets the shortcut target for the given word. Returns null if the word is not in the 196 * shortcut list. 197 */ 198 public WeightedString getShortcut(final String word) { 199 // TODO: Don't do a linear search 200 if (mShortcutTargets != null) { 201 final int size = mShortcutTargets.size(); 202 for (int i = 0; i < size; ++i) { 203 WeightedString shortcut = mShortcutTargets.get(i); 204 if (shortcut.mWord.equals(word)) { 205 return shortcut; 206 } 207 } 208 } 209 return null; 210 } 211 212 /** 213 * Gets the bigram for the given word. 214 * Returns null if the word is not in the bigrams list. 215 */ 216 public WeightedString getBigram(final String word) { 217 // TODO: Don't do a linear search 218 if (mBigrams != null) { 219 final int size = mBigrams.size(); 220 for (int i = 0; i < size; ++i) { 221 WeightedString bigram = mBigrams.get(i); 222 if (bigram.mWord.equals(word)) { 223 return bigram; 224 } 225 } 226 } 227 return null; 228 } 229 230 /** 231 * Updates the CharGroup with the given properties. Adds the shortcut and bigram lists to 232 * the existing ones if any. Note: unigram, bigram, and shortcut frequencies are only 233 * updated if they are higher than the existing ones. 234 */ 235 public void update(final int frequency, final ArrayList<WeightedString> shortcutTargets, 236 final ArrayList<WeightedString> bigrams, 237 final boolean isNotAWord, final boolean isBlacklistEntry) { 238 if (frequency > mFrequency) { 239 mFrequency = frequency; 240 } 241 if (shortcutTargets != null) { 242 if (mShortcutTargets == null) { 243 mShortcutTargets = shortcutTargets; 244 } else { 245 final int size = shortcutTargets.size(); 246 for (int i = 0; i < size; ++i) { 247 final WeightedString shortcut = shortcutTargets.get(i); 248 final WeightedString existingShortcut = getShortcut(shortcut.mWord); 249 if (existingShortcut == null) { 250 mShortcutTargets.add(shortcut); 251 } else if (existingShortcut.mFrequency < shortcut.mFrequency) { 252 existingShortcut.mFrequency = shortcut.mFrequency; 253 } 254 } 255 } 256 } 257 if (bigrams != null) { 258 if (mBigrams == null) { 259 mBigrams = bigrams; 260 } else { 261 final int size = bigrams.size(); 262 for (int i = 0; i < size; ++i) { 263 final WeightedString bigram = bigrams.get(i); 264 final WeightedString existingBigram = getBigram(bigram.mWord); 265 if (existingBigram == null) { 266 mBigrams.add(bigram); 267 } else if (existingBigram.mFrequency < bigram.mFrequency) { 268 existingBigram.mFrequency = bigram.mFrequency; 269 } 270 } 271 } 272 } 273 mIsNotAWord = isNotAWord; 274 mIsBlacklistEntry = isBlacklistEntry; 275 } 276 } 277 278 /** 279 * Options global to the dictionary. 280 */ 281 public static final class DictionaryOptions { 282 public final boolean mGermanUmlautProcessing; 283 public final boolean mFrenchLigatureProcessing; 284 public final HashMap<String, String> mAttributes; 285 public DictionaryOptions(final HashMap<String, String> attributes, 286 final boolean germanUmlautProcessing, final boolean frenchLigatureProcessing) { 287 mAttributes = attributes; 288 mGermanUmlautProcessing = germanUmlautProcessing; 289 mFrenchLigatureProcessing = frenchLigatureProcessing; 290 } 291 @Override 292 public String toString() { // Convenience method 293 return toString(0, false); 294 } 295 public String toString(final int indentCount, final boolean plumbing) { 296 final StringBuilder indent = new StringBuilder(); 297 if (plumbing) { 298 indent.append("H:"); 299 } else { 300 for (int i = 0; i < indentCount; ++i) { 301 indent.append(" "); 302 } 303 } 304 final StringBuilder s = new StringBuilder(); 305 for (final String optionKey : mAttributes.keySet()) { 306 s.append(indent); 307 s.append(optionKey); 308 s.append(" = "); 309 if ("date".equals(optionKey) && !plumbing) { 310 // Date needs a number of milliseconds, but the dictionary contains seconds 311 s.append(new Date( 312 1000 * Long.parseLong(mAttributes.get(optionKey))).toString()); 313 } else { 314 s.append(mAttributes.get(optionKey)); 315 } 316 s.append("\n"); 317 } 318 if (mGermanUmlautProcessing) { 319 s.append(indent); 320 s.append("Needs German umlaut processing\n"); 321 } 322 if (mFrenchLigatureProcessing) { 323 s.append(indent); 324 s.append("Needs French ligature processing\n"); 325 } 326 return s.toString(); 327 } 328 } 329 330 public final DictionaryOptions mOptions; 331 public final Node mRoot; 332 333 public FusionDictionary(final Node root, final DictionaryOptions options) { 334 mRoot = root; 335 mOptions = options; 336 } 337 338 public void addOptionAttribute(final String key, final String value) { 339 mOptions.mAttributes.put(key, value); 340 } 341 342 /** 343 * Helper method to convert a String to an int array. 344 */ 345 static int[] getCodePoints(final String word) { 346 // TODO: this is a copy-paste of the contents of StringUtils.toCodePointArray, 347 // which is not visible from the makedict package. Factor this code. 348 final char[] characters = word.toCharArray(); 349 final int length = characters.length; 350 final int[] codePoints = new int[Character.codePointCount(characters, 0, length)]; 351 int codePoint = Character.codePointAt(characters, 0); 352 int dsti = 0; 353 for (int srci = Character.charCount(codePoint); 354 srci < length; srci += Character.charCount(codePoint), ++dsti) { 355 codePoints[dsti] = codePoint; 356 codePoint = Character.codePointAt(characters, srci); 357 } 358 codePoints[dsti] = codePoint; 359 return codePoints; 360 } 361 362 /** 363 * Helper method to add a word as a string. 364 * 365 * This method adds a word to the dictionary with the given frequency. Optional 366 * lists of bigrams and shortcuts can be passed here. For each word inside, 367 * they will be added to the dictionary as necessary. 368 * 369 * @param word the word to add. 370 * @param frequency the frequency of the word, in the range [0..255]. 371 * @param shortcutTargets a list of shortcut targets for this word, or null. 372 * @param isNotAWord true if this should not be considered a word (e.g. shortcut only) 373 */ 374 public void add(final String word, final int frequency, 375 final ArrayList<WeightedString> shortcutTargets, final boolean isNotAWord) { 376 add(getCodePoints(word), frequency, shortcutTargets, isNotAWord, 377 false /* isBlacklistEntry */); 378 } 379 380 /** 381 * Helper method to add a blacklist entry as a string. 382 * 383 * @param word the word to add as a blacklist entry. 384 * @param shortcutTargets a list of shortcut targets for this word, or null. 385 * @param isNotAWord true if this is not a word for spellcheking purposes (shortcut only or so) 386 */ 387 public void addBlacklistEntry(final String word, 388 final ArrayList<WeightedString> shortcutTargets, final boolean isNotAWord) { 389 add(getCodePoints(word), 0, shortcutTargets, isNotAWord, true /* isBlacklistEntry */); 390 } 391 392 /** 393 * Sanity check for a node. 394 * 395 * This method checks that all CharGroups in a node are ordered as expected. 396 * If they are, nothing happens. If they aren't, an exception is thrown. 397 */ 398 private void checkStack(Node node) { 399 ArrayList<CharGroup> stack = node.mData; 400 int lastValue = -1; 401 for (int i = 0; i < stack.size(); ++i) { 402 int currentValue = stack.get(i).mChars[0]; 403 if (currentValue <= lastValue) 404 throw new RuntimeException("Invalid stack"); 405 else 406 lastValue = currentValue; 407 } 408 } 409 410 /** 411 * Helper method to add a new bigram to the dictionary. 412 * 413 * @param word1 the previous word of the context 414 * @param word2 the next word of the context 415 * @param frequency the bigram frequency 416 */ 417 public void setBigram(final String word1, final String word2, final int frequency) { 418 CharGroup charGroup = findWordInTree(mRoot, word1); 419 if (charGroup != null) { 420 final CharGroup charGroup2 = findWordInTree(mRoot, word2); 421 if (charGroup2 == null) { 422 add(getCodePoints(word2), 0, null, false /* isNotAWord */, 423 false /* isBlacklistEntry */); 424 // The chargroup for the first word may have moved by the above insertion, 425 // if word1 and word2 share a common stem that happens not to have been 426 // a cutting point until now. In this case, we need to refresh charGroup. 427 charGroup = findWordInTree(mRoot, word1); 428 } 429 charGroup.addBigram(word2, frequency); 430 } else { 431 throw new RuntimeException("First word of bigram not found"); 432 } 433 } 434 435 /** 436 * Add a word to this dictionary. 437 * 438 * The shortcuts, if any, have to be in the dictionary already. If they aren't, 439 * an exception is thrown. 440 * 441 * @param word the word, as an int array. 442 * @param frequency the frequency of the word, in the range [0..255]. 443 * @param shortcutTargets an optional list of shortcut targets for this word (null if none). 444 * @param isNotAWord true if this is not a word for spellcheking purposes (shortcut only or so) 445 * @param isBlacklistEntry true if this is a blacklisted word, false otherwise 446 */ 447 private void add(final int[] word, final int frequency, 448 final ArrayList<WeightedString> shortcutTargets, 449 final boolean isNotAWord, final boolean isBlacklistEntry) { 450 assert(frequency >= 0 && frequency <= 255); 451 if (word.length >= Constants.Dictionary.MAX_WORD_LENGTH) { 452 MakedictLog.w("Ignoring a word that is too long: word.length = " + word.length); 453 return; 454 } 455 456 Node currentNode = mRoot; 457 int charIndex = 0; 458 459 CharGroup currentGroup = null; 460 int differentCharIndex = 0; // Set by the loop to the index of the char that differs 461 int nodeIndex = findIndexOfChar(mRoot, word[charIndex]); 462 while (CHARACTER_NOT_FOUND != nodeIndex) { 463 currentGroup = currentNode.mData.get(nodeIndex); 464 differentCharIndex = compareArrays(currentGroup.mChars, word, charIndex); 465 if (ARRAYS_ARE_EQUAL != differentCharIndex 466 && differentCharIndex < currentGroup.mChars.length) break; 467 if (null == currentGroup.mChildren) break; 468 charIndex += currentGroup.mChars.length; 469 if (charIndex >= word.length) break; 470 currentNode = currentGroup.mChildren; 471 nodeIndex = findIndexOfChar(currentNode, word[charIndex]); 472 } 473 474 if (-1 == nodeIndex) { 475 // No node at this point to accept the word. Create one. 476 final int insertionIndex = findInsertionIndex(currentNode, word[charIndex]); 477 final CharGroup newGroup = new CharGroup( 478 Arrays.copyOfRange(word, charIndex, word.length), 479 shortcutTargets, null /* bigrams */, frequency, isNotAWord, isBlacklistEntry); 480 currentNode.mData.add(insertionIndex, newGroup); 481 if (DBG) checkStack(currentNode); 482 } else { 483 // There is a word with a common prefix. 484 if (differentCharIndex == currentGroup.mChars.length) { 485 if (charIndex + differentCharIndex >= word.length) { 486 // The new word is a prefix of an existing word, but the node on which it 487 // should end already exists as is. Since the old CharNode was not a terminal, 488 // make it one by filling in its frequency and other attributes 489 currentGroup.update(frequency, shortcutTargets, null, isNotAWord, 490 isBlacklistEntry); 491 } else { 492 // The new word matches the full old word and extends past it. 493 // We only have to create a new node and add it to the end of this. 494 final CharGroup newNode = new CharGroup( 495 Arrays.copyOfRange(word, charIndex + differentCharIndex, word.length), 496 shortcutTargets, null /* bigrams */, frequency, isNotAWord, 497 isBlacklistEntry); 498 currentGroup.mChildren = new Node(); 499 currentGroup.mChildren.mData.add(newNode); 500 } 501 } else { 502 if (0 == differentCharIndex) { 503 // Exact same word. Update the frequency if higher. This will also add the 504 // new shortcuts to the existing shortcut list if it already exists. 505 currentGroup.update(frequency, shortcutTargets, null, 506 currentGroup.mIsNotAWord && isNotAWord, 507 currentGroup.mIsBlacklistEntry || isBlacklistEntry); 508 } else { 509 // Partial prefix match only. We have to replace the current node with a node 510 // containing the current prefix and create two new ones for the tails. 511 Node newChildren = new Node(); 512 final CharGroup newOldWord = new CharGroup( 513 Arrays.copyOfRange(currentGroup.mChars, differentCharIndex, 514 currentGroup.mChars.length), currentGroup.mShortcutTargets, 515 currentGroup.mBigrams, currentGroup.mFrequency, 516 currentGroup.mIsNotAWord, currentGroup.mIsBlacklistEntry, 517 currentGroup.mChildren); 518 newChildren.mData.add(newOldWord); 519 520 final CharGroup newParent; 521 if (charIndex + differentCharIndex >= word.length) { 522 newParent = new CharGroup( 523 Arrays.copyOfRange(currentGroup.mChars, 0, differentCharIndex), 524 shortcutTargets, null /* bigrams */, frequency, 525 isNotAWord, isBlacklistEntry, newChildren); 526 } else { 527 newParent = new CharGroup( 528 Arrays.copyOfRange(currentGroup.mChars, 0, differentCharIndex), 529 null /* shortcutTargets */, null /* bigrams */, -1, 530 false /* isNotAWord */, false /* isBlacklistEntry */, newChildren); 531 final CharGroup newWord = new CharGroup(Arrays.copyOfRange(word, 532 charIndex + differentCharIndex, word.length), 533 shortcutTargets, null /* bigrams */, frequency, 534 isNotAWord, isBlacklistEntry); 535 final int addIndex = word[charIndex + differentCharIndex] 536 > currentGroup.mChars[differentCharIndex] ? 1 : 0; 537 newChildren.mData.add(addIndex, newWord); 538 } 539 currentNode.mData.set(nodeIndex, newParent); 540 } 541 if (DBG) checkStack(currentNode); 542 } 543 } 544 } 545 546 private static int ARRAYS_ARE_EQUAL = 0; 547 548 /** 549 * Custom comparison of two int arrays taken to contain character codes. 550 * 551 * This method compares the two arrays passed as an argument in a lexicographic way, 552 * with an offset in the dst string. 553 * This method does NOT test for the first character. It is taken to be equal. 554 * I repeat: this method starts the comparison at 1 <> dstOffset + 1. 555 * The index where the strings differ is returned. ARRAYS_ARE_EQUAL = 0 is returned if the 556 * strings are equal. This works BECAUSE we don't look at the first character. 557 * 558 * @param src the left-hand side string of the comparison. 559 * @param dst the right-hand side string of the comparison. 560 * @param dstOffset the offset in the right-hand side string. 561 * @return the index at which the strings differ, or ARRAYS_ARE_EQUAL = 0 if they don't. 562 */ 563 private static int compareArrays(final int[] src, final int[] dst, int dstOffset) { 564 // We do NOT test the first char, because we come from a method that already 565 // tested it. 566 for (int i = 1; i < src.length; ++i) { 567 if (dstOffset + i >= dst.length) return i; 568 if (src[i] != dst[dstOffset + i]) return i; 569 } 570 if (dst.length > src.length) return src.length; 571 return ARRAYS_ARE_EQUAL; 572 } 573 574 /** 575 * Helper class that compares and sorts two chargroups according to their 576 * first element only. I repeat: ONLY the first element is considered, the rest 577 * is ignored. 578 * This comparator imposes orderings that are inconsistent with equals. 579 */ 580 static private final class CharGroupComparator implements java.util.Comparator<CharGroup> { 581 @Override 582 public int compare(CharGroup c1, CharGroup c2) { 583 if (c1.mChars[0] == c2.mChars[0]) return 0; 584 return c1.mChars[0] < c2.mChars[0] ? -1 : 1; 585 } 586 } 587 final static private CharGroupComparator CHARGROUP_COMPARATOR = new CharGroupComparator(); 588 589 /** 590 * Finds the insertion index of a character within a node. 591 */ 592 private static int findInsertionIndex(final Node node, int character) { 593 final ArrayList<CharGroup> data = node.mData; 594 final CharGroup reference = new CharGroup(new int[] { character }, 595 null /* shortcutTargets */, null /* bigrams */, 0, false /* isNotAWord */, 596 false /* isBlacklistEntry */); 597 int result = Collections.binarySearch(data, reference, CHARGROUP_COMPARATOR); 598 return result >= 0 ? result : -result - 1; 599 } 600 601 private static int CHARACTER_NOT_FOUND = -1; 602 603 /** 604 * Find the index of a char in a node, if it exists. 605 * 606 * @param node the node to search in. 607 * @param character the character to search for. 608 * @return the position of the character if it's there, or CHARACTER_NOT_FOUND = -1 else. 609 */ 610 private static int findIndexOfChar(final Node node, int character) { 611 final int insertionIndex = findInsertionIndex(node, character); 612 if (node.mData.size() <= insertionIndex) return CHARACTER_NOT_FOUND; 613 return character == node.mData.get(insertionIndex).mChars[0] ? insertionIndex 614 : CHARACTER_NOT_FOUND; 615 } 616 617 /** 618 * Helper method to find a word in a given branch. 619 */ 620 public static CharGroup findWordInTree(Node node, final String s) { 621 int index = 0; 622 final StringBuilder checker = DBG ? new StringBuilder() : null; 623 624 CharGroup currentGroup; 625 final int codePointCountInS = s.codePointCount(0, s.length()); 626 do { 627 int indexOfGroup = findIndexOfChar(node, s.codePointAt(index)); 628 if (CHARACTER_NOT_FOUND == indexOfGroup) return null; 629 currentGroup = node.mData.get(indexOfGroup); 630 631 if (s.length() - index < currentGroup.mChars.length) return null; 632 int newIndex = index; 633 while (newIndex < s.length() && newIndex - index < currentGroup.mChars.length) { 634 if (currentGroup.mChars[newIndex - index] != s.codePointAt(newIndex)) return null; 635 newIndex++; 636 } 637 index = newIndex; 638 639 if (DBG) checker.append(new String(currentGroup.mChars, 0, currentGroup.mChars.length)); 640 if (index < codePointCountInS) { 641 node = currentGroup.mChildren; 642 } 643 } while (null != node && index < codePointCountInS); 644 645 if (index < codePointCountInS) return null; 646 if (!currentGroup.isTerminal()) return null; 647 if (DBG && !s.equals(checker.toString())) return null; 648 return currentGroup; 649 } 650 651 /** 652 * Helper method to find out whether a word is in the dict or not. 653 */ 654 public boolean hasWord(final String s) { 655 if (null == s || "".equals(s)) { 656 throw new RuntimeException("Can't search for a null or empty string"); 657 } 658 return null != findWordInTree(mRoot, s); 659 } 660 661 /** 662 * Recursively count the number of character groups in a given branch of the trie. 663 * 664 * @param node the parent node. 665 * @return the number of char groups in all the branch under this node. 666 */ 667 public static int countCharGroups(final Node node) { 668 final int nodeSize = node.mData.size(); 669 int size = nodeSize; 670 for (int i = nodeSize - 1; i >= 0; --i) { 671 CharGroup group = node.mData.get(i); 672 if (null != group.mChildren) 673 size += countCharGroups(group.mChildren); 674 } 675 return size; 676 } 677 678 /** 679 * Recursively count the number of nodes in a given branch of the trie. 680 * 681 * @param node the node to count. 682 * @return the number of nodes in this branch. 683 */ 684 public static int countNodes(final Node node) { 685 int size = 1; 686 for (int i = node.mData.size() - 1; i >= 0; --i) { 687 CharGroup group = node.mData.get(i); 688 if (null != group.mChildren) 689 size += countNodes(group.mChildren); 690 } 691 return size; 692 } 693 694 // Recursively find out whether there are any bigrams. 695 // This can be pretty expensive especially if there aren't any (we return as soon 696 // as we find one, so it's much cheaper if there are bigrams) 697 private static boolean hasBigramsInternal(final Node node) { 698 if (null == node) return false; 699 for (int i = node.mData.size() - 1; i >= 0; --i) { 700 CharGroup group = node.mData.get(i); 701 if (null != group.mBigrams) return true; 702 if (hasBigramsInternal(group.mChildren)) return true; 703 } 704 return false; 705 } 706 707 /** 708 * Finds out whether there are any bigrams in this dictionary. 709 * 710 * @return true if there is any bigram, false otherwise. 711 */ 712 // TODO: this is expensive especially for large dictionaries without any bigram. 713 // The up side is, this is always accurate and correct and uses no memory. We should 714 // find a more efficient way of doing this, without compromising too much on memory 715 // and ease of use. 716 public boolean hasBigrams() { 717 return hasBigramsInternal(mRoot); 718 } 719 720 // Historically, the tails of the words were going to be merged to save space. 721 // However, that would prevent the code to search for a specific address in log(n) 722 // time so this was abandoned. 723 // The code is still of interest as it does add some compression to any dictionary 724 // that has no need for attributes. Implementations that does not read attributes should be 725 // able to read a dictionary with merged tails. 726 // Also, the following code does support frequencies, as in, it will only merges 727 // tails that share the same frequency. Though it would result in the above loss of 728 // performance while searching by address, it is still technically possible to merge 729 // tails that contain attributes, but this code does not take that into account - it does 730 // not compare attributes and will merge terminals with different attributes regardless. 731 public void mergeTails() { 732 MakedictLog.i("Do not merge tails"); 733 return; 734 735// MakedictLog.i("Merging nodes. Number of nodes : " + countNodes(root)); 736// MakedictLog.i("Number of groups : " + countCharGroups(root)); 737// 738// final HashMap<String, ArrayList<Node>> repository = 739// new HashMap<String, ArrayList<Node>>(); 740// mergeTailsInner(repository, root); 741// 742// MakedictLog.i("Number of different pseudohashes : " + repository.size()); 743// int size = 0; 744// for (ArrayList<Node> a : repository.values()) { 745// size += a.size(); 746// } 747// MakedictLog.i("Number of nodes after merge : " + (1 + size)); 748// MakedictLog.i("Recursively seen nodes : " + countNodes(root)); 749 } 750 751 // The following methods are used by the deactivated mergeTails() 752// private static boolean isEqual(Node a, Node b) { 753// if (null == a && null == b) return true; 754// if (null == a || null == b) return false; 755// if (a.data.size() != b.data.size()) return false; 756// final int size = a.data.size(); 757// for (int i = size - 1; i >= 0; --i) { 758// CharGroup aGroup = a.data.get(i); 759// CharGroup bGroup = b.data.get(i); 760// if (aGroup.frequency != bGroup.frequency) return false; 761// if (aGroup.alternates == null && bGroup.alternates != null) return false; 762// if (aGroup.alternates != null && !aGroup.equals(bGroup.alternates)) return false; 763// if (!Arrays.equals(aGroup.chars, bGroup.chars)) return false; 764// if (!isEqual(aGroup.children, bGroup.children)) return false; 765// } 766// return true; 767// } 768 769// static private HashMap<String, ArrayList<Node>> mergeTailsInner( 770// final HashMap<String, ArrayList<Node>> map, final Node node) { 771// final ArrayList<CharGroup> branches = node.data; 772// final int nodeSize = branches.size(); 773// for (int i = 0; i < nodeSize; ++i) { 774// CharGroup group = branches.get(i); 775// if (null != group.children) { 776// String pseudoHash = getPseudoHash(group.children); 777// ArrayList<Node> similarList = map.get(pseudoHash); 778// if (null == similarList) { 779// similarList = new ArrayList<Node>(); 780// map.put(pseudoHash, similarList); 781// } 782// boolean merged = false; 783// for (Node similar : similarList) { 784// if (isEqual(group.children, similar)) { 785// group.children = similar; 786// merged = true; 787// break; 788// } 789// } 790// if (!merged) { 791// similarList.add(group.children); 792// } 793// mergeTailsInner(map, group.children); 794// } 795// } 796// return map; 797// } 798 799// private static String getPseudoHash(final Node node) { 800// StringBuilder s = new StringBuilder(); 801// for (CharGroup g : node.data) { 802// s.append(g.frequency); 803// for (int ch : g.chars) { 804// s.append(Character.toChars(ch)); 805// } 806// } 807// return s.toString(); 808// } 809 810 /** 811 * Iterator to walk through a dictionary. 812 * 813 * This is purely for convenience. 814 */ 815 public static final class DictionaryIterator implements Iterator<Word> { 816 private static final class Position { 817 public Iterator<CharGroup> pos; 818 public int length; 819 public Position(ArrayList<CharGroup> groups) { 820 pos = groups.iterator(); 821 length = 0; 822 } 823 } 824 final StringBuilder mCurrentString; 825 final LinkedList<Position> mPositions; 826 827 public DictionaryIterator(ArrayList<CharGroup> root) { 828 mCurrentString = new StringBuilder(); 829 mPositions = new LinkedList<Position>(); 830 final Position rootPos = new Position(root); 831 mPositions.add(rootPos); 832 } 833 834 @Override 835 public boolean hasNext() { 836 for (Position p : mPositions) { 837 if (p.pos.hasNext()) { 838 return true; 839 } 840 } 841 return false; 842 } 843 844 @Override 845 public Word next() { 846 Position currentPos = mPositions.getLast(); 847 mCurrentString.setLength(mCurrentString.length() - currentPos.length); 848 849 do { 850 if (currentPos.pos.hasNext()) { 851 final CharGroup currentGroup = currentPos.pos.next(); 852 currentPos.length = currentGroup.mChars.length; 853 for (int i : currentGroup.mChars) 854 mCurrentString.append(Character.toChars(i)); 855 if (null != currentGroup.mChildren) { 856 currentPos = new Position(currentGroup.mChildren.mData); 857 mPositions.addLast(currentPos); 858 } 859 if (currentGroup.mFrequency >= 0) 860 return new Word(mCurrentString.toString(), currentGroup.mFrequency, 861 currentGroup.mShortcutTargets, currentGroup.mBigrams, 862 currentGroup.mIsNotAWord, currentGroup.mIsBlacklistEntry); 863 } else { 864 mPositions.removeLast(); 865 currentPos = mPositions.getLast(); 866 mCurrentString.setLength(mCurrentString.length() - mPositions.getLast().length); 867 } 868 } while (true); 869 } 870 871 @Override 872 public void remove() { 873 throw new UnsupportedOperationException("Unsupported yet"); 874 } 875 876 } 877 878 /** 879 * Method to return an iterator. 880 * 881 * This method enables Java's enhanced for loop. With this you can have a FusionDictionary x 882 * and say : for (Word w : x) {} 883 */ 884 @Override 885 public Iterator<Word> iterator() { 886 return new DictionaryIterator(mRoot.mData); 887 } 888} 889