Merge "Move entry updating method to language model dict content."

main
Keisuke Kuroyanagi 2014-10-15 04:51:03 +00:00 committed by Android (Google) Code Review
commit c2429c54ac
7 changed files with 116 additions and 73 deletions

View File

@ -25,6 +25,7 @@ namespace latinime {
const int LanguageModelDictContent::UNIGRAM_COUNT_INDEX_IN_ENTRY_COUNT_TABLE = 0;
const int LanguageModelDictContent::BIGRAM_COUNT_INDEX_IN_ENTRY_COUNT_TABLE = 1;
const int LanguageModelDictContent::DUMMY_PROBABILITY_FOR_VALID_WORDS = 1;
bool LanguageModelDictContent::save(FILE *const file) const {
return mTrieMap.save(file);
@ -143,6 +144,56 @@ bool LanguageModelDictContent::truncateEntries(const int *const entryCounts,
return true;
}
bool LanguageModelDictContent::updateAllEntriesOnInputWord(const WordIdArrayView prevWordIds,
const int wordId, const bool isValid, const HistoricalInfo historicalInfo,
const HeaderPolicy *const headerPolicy, int *const outAddedNewNgramEntryCount) {
if (outAddedNewNgramEntryCount) {
*outAddedNewNgramEntryCount = 0;
}
if (!mHasHistoricalInfo) {
AKLOGE("updateAllEntriesOnInputWord is called for dictionary without historical info.");
return false;
}
const ProbabilityEntry originalUnigramProbabilityEntry = getProbabilityEntry(wordId);
const ProbabilityEntry updatedUnigramProbabilityEntry = createUpdatedEntryFrom(
originalUnigramProbabilityEntry, isValid, historicalInfo, headerPolicy);
if (!setProbabilityEntry(wordId, &updatedUnigramProbabilityEntry)) {
return false;
}
for (size_t i = 0; i < prevWordIds.size(); ++i) {
if (prevWordIds[i] == NOT_A_WORD_ID) {
break;
}
// TODO: Optimize this code.
const WordIdArrayView limitedPrevWordIds = prevWordIds.limit(i + 1);
const ProbabilityEntry originalNgramProbabilityEntry = getNgramProbabilityEntry(
limitedPrevWordIds, wordId);
const ProbabilityEntry updatedNgramProbabilityEntry = createUpdatedEntryFrom(
originalNgramProbabilityEntry, isValid, historicalInfo, headerPolicy);
if (!setNgramProbabilityEntry(limitedPrevWordIds, wordId, &updatedNgramProbabilityEntry)) {
return false;
}
if (!originalNgramProbabilityEntry.isValid() && outAddedNewNgramEntryCount) {
*outAddedNewNgramEntryCount += 1;
}
}
return true;
}
const ProbabilityEntry LanguageModelDictContent::createUpdatedEntryFrom(
const ProbabilityEntry &originalProbabilityEntry, const bool isValid,
const HistoricalInfo historicalInfo, const HeaderPolicy *const headerPolicy) const {
const HistoricalInfo updatedHistoricalInfo = ForgettingCurveUtils::createUpdatedHistoricalInfo(
originalProbabilityEntry.getHistoricalInfo(), isValid ?
DUMMY_PROBABILITY_FOR_VALID_WORDS : NOT_A_PROBABILITY,
&historicalInfo, headerPolicy);
if (originalProbabilityEntry.isValid()) {
return ProbabilityEntry(originalProbabilityEntry.getFlags(), &updatedHistoricalInfo);
} else {
return ProbabilityEntry(0 /* flags */, &updatedHistoricalInfo);
}
}
bool LanguageModelDictContent::runGCInner(
const TerminalPositionLookupTable::TerminalIdMap *const terminalIdMap,
const TrieMap::TrieMapRange trieMapRange,
@ -203,7 +254,7 @@ int LanguageModelDictContent::getBitmapEntryIndex(const WordIdArrayView prevWord
return bitmapEntryIndex;
}
bool LanguageModelDictContent::updateAllProbabilityEntriesInner(const int bitmapEntryIndex,
bool LanguageModelDictContent::updateAllProbabilityEntriesForGCInner(const int bitmapEntryIndex,
const int level, const HeaderPolicy *const headerPolicy, int *const outEntryCounts) {
for (const auto &entry : mTrieMap.getEntriesInSpecifiedLevel(bitmapEntryIndex)) {
if (level > MAX_PREV_WORD_COUNT_FOR_N_GRAM) {
@ -237,7 +288,7 @@ bool LanguageModelDictContent::updateAllProbabilityEntriesInner(const int bitmap
if (!entry.hasNextLevelMap()) {
continue;
}
if (!updateAllProbabilityEntriesInner(entry.getNextLevelBitmapEntryIndex(), level + 1,
if (!updateAllProbabilityEntriesForGCInner(entry.getNextLevelBitmapEntryIndex(), level + 1,
headerPolicy, outEntryCounts)) {
return false;
}

View File

@ -154,19 +154,23 @@ class LanguageModelDictContent {
EntryRange getProbabilityEntries(const WordIdArrayView prevWordIds) const;
bool updateAllProbabilityEntries(const HeaderPolicy *const headerPolicy,
bool updateAllProbabilityEntriesForGC(const HeaderPolicy *const headerPolicy,
int *const outEntryCounts) {
for (int i = 0; i <= MAX_PREV_WORD_COUNT_FOR_N_GRAM; ++i) {
outEntryCounts[i] = 0;
}
return updateAllProbabilityEntriesInner(mTrieMap.getRootBitmapEntryIndex(), 0 /* level */,
headerPolicy, outEntryCounts);
return updateAllProbabilityEntriesForGCInner(mTrieMap.getRootBitmapEntryIndex(),
0 /* level */, headerPolicy, outEntryCounts);
}
// entryCounts should be created by updateAllProbabilityEntries.
bool truncateEntries(const int *const entryCounts, const int *const maxEntryCounts,
const HeaderPolicy *const headerPolicy, int *const outEntryCounts);
bool updateAllEntriesOnInputWord(const WordIdArrayView prevWordIds, const int wordId,
const bool isValid, const HistoricalInfo historicalInfo,
const HeaderPolicy *const headerPolicy, int *const outAddedNewNgramEntryCount);
private:
DISALLOW_COPY_AND_ASSIGN(LanguageModelDictContent);
@ -193,6 +197,9 @@ class LanguageModelDictContent {
DISALLOW_DEFAULT_CONSTRUCTOR(EntryInfoToTurncate);
};
// TODO: Remove
static const int DUMMY_PROBABILITY_FOR_VALID_WORDS;
TrieMap mTrieMap;
const bool mHasHistoricalInfo;
@ -201,13 +208,16 @@ class LanguageModelDictContent {
int *const outNgramCount);
int createAndGetBitmapEntryIndex(const WordIdArrayView prevWordIds);
int getBitmapEntryIndex(const WordIdArrayView prevWordIds) const;
bool updateAllProbabilityEntriesInner(const int bitmapEntryIndex, const int level,
bool updateAllProbabilityEntriesForGCInner(const int bitmapEntryIndex, const int level,
const HeaderPolicy *const headerPolicy, int *const outEntryCounts);
bool turncateEntriesInSpecifiedLevel(const HeaderPolicy *const headerPolicy,
const int maxEntryCount, const int targetLevel, int *const outEntryCount);
bool getEntryInfo(const HeaderPolicy *const headerPolicy, const int targetLevel,
const int bitmapEntryIndex, std::vector<int> *const prevWordIds,
std::vector<EntryInfoToTurncate> *const outEntryInfo) const;
const ProbabilityEntry createUpdatedEntryFrom(const ProbabilityEntry &originalProbabilityEntry,
const bool isValid, const HistoricalInfo historicalInfo,
const HeaderPolicy *const headerPolicy) const;
};
} // namespace latinime
#endif /* LATINIME_LANGUAGE_MODEL_DICT_CONTENT_H */

View File

@ -142,14 +142,9 @@ bool Ver4PatriciaTrieNodeWriter::updatePtNodeUnigramProperty(
if (!toBeUpdatedPtNodeParams->isTerminal()) {
return false;
}
const ProbabilityEntry originalProbabilityEntry =
mBuffers->getLanguageModelDictContent()->getProbabilityEntry(
toBeUpdatedPtNodeParams->getTerminalId());
const ProbabilityEntry probabilityEntryOfUnigramProperty = ProbabilityEntry(unigramProperty);
const ProbabilityEntry updatedProbabilityEntry =
createUpdatedEntryFrom(&originalProbabilityEntry, &probabilityEntryOfUnigramProperty);
return mBuffers->getMutableLanguageModelDictContent()->setProbabilityEntry(
toBeUpdatedPtNodeParams->getTerminalId(), &updatedProbabilityEntry);
toBeUpdatedPtNodeParams->getTerminalId(), &probabilityEntryOfUnigramProperty);
}
bool Ver4PatriciaTrieNodeWriter::updatePtNodeProbabilityAndGetNeedsToKeepPtNodeAfterGC(
@ -203,10 +198,8 @@ bool Ver4PatriciaTrieNodeWriter::writeNewTerminalPtNodeAndAdvancePosition(
// Write probability.
ProbabilityEntry newProbabilityEntry;
const ProbabilityEntry probabilityEntryOfUnigramProperty = ProbabilityEntry(unigramProperty);
const ProbabilityEntry probabilityEntryToWrite = createUpdatedEntryFrom(
&newProbabilityEntry, &probabilityEntryOfUnigramProperty);
return mBuffers->getMutableLanguageModelDictContent()->setProbabilityEntry(
terminalId, &probabilityEntryToWrite);
terminalId, &probabilityEntryOfUnigramProperty);
}
// TODO: Support counting ngram entries.
@ -217,10 +210,8 @@ bool Ver4PatriciaTrieNodeWriter::addNgramEntry(const WordIdArrayView prevWordIds
const ProbabilityEntry probabilityEntry =
languageModelDictContent->getNgramProbabilityEntry(prevWordIds, wordId);
const ProbabilityEntry probabilityEntryOfNgramProperty(ngramProperty);
const ProbabilityEntry updatedProbabilityEntry = createUpdatedEntryFrom(
&probabilityEntry, &probabilityEntryOfNgramProperty);
if (!languageModelDictContent->setNgramProbabilityEntry(
prevWordIds, wordId, &updatedProbabilityEntry)) {
prevWordIds, wordId, &probabilityEntryOfNgramProperty)) {
AKLOGE("Cannot add new ngram entry. prevWordId[0]: %d, prevWordId.size(): %zd, wordId: %d",
prevWordIds[0], prevWordIds.size(), wordId);
return false;
@ -346,22 +337,6 @@ bool Ver4PatriciaTrieNodeWriter::writePtNodeAndGetTerminalIdAndAdvancePosition(
ptNodeParams->getCodePointCount() > 1 /* hasMultipleChars */);
}
// TODO: Move probability handling code to LanguageModelDictContent.
const ProbabilityEntry Ver4PatriciaTrieNodeWriter::createUpdatedEntryFrom(
const ProbabilityEntry *const originalProbabilityEntry,
const ProbabilityEntry *const probabilityEntry) const {
if (mHeaderPolicy->hasHistoricalInfoOfWords()) {
const HistoricalInfo updatedHistoricalInfo =
ForgettingCurveUtils::createUpdatedHistoricalInfo(
originalProbabilityEntry->getHistoricalInfo(),
probabilityEntry->getProbability(), probabilityEntry->getHistoricalInfo(),
mHeaderPolicy);
return ProbabilityEntry(probabilityEntry->getFlags(), &updatedHistoricalInfo);
} else {
return *probabilityEntry;
}
}
bool Ver4PatriciaTrieNodeWriter::updatePtNodeFlags(const int ptNodePos, const bool isTerminal,
const bool hasMultipleChars) {
// Create node flags and write them.

View File

@ -38,11 +38,10 @@ class Ver4ShortcutListPolicy;
class Ver4PatriciaTrieNodeWriter : public PtNodeWriter {
public:
Ver4PatriciaTrieNodeWriter(BufferWithExtendableBuffer *const trieBuffer,
Ver4DictBuffers *const buffers, const HeaderPolicy *const headerPolicy,
const PtNodeReader *const ptNodeReader,
Ver4DictBuffers *const buffers, const PtNodeReader *const ptNodeReader,
const PtNodeArrayReader *const ptNodeArrayReader,
Ver4ShortcutListPolicy *const shortcutPolicy)
: mTrieBuffer(trieBuffer), mBuffers(buffers), mHeaderPolicy(headerPolicy),
: mTrieBuffer(trieBuffer), mBuffers(buffers),
mReadingHelper(ptNodeReader, ptNodeArrayReader), mShortcutPolicy(shortcutPolicy) {}
virtual ~Ver4PatriciaTrieNodeWriter() {}
@ -96,20 +95,12 @@ class Ver4PatriciaTrieNodeWriter : public PtNodeWriter {
const PtNodeParams *const ptNodeParams, int *const outTerminalId,
int *const ptNodeWritingPos);
// Create updated probability entry using given probability property. In addition to the
// probability, this method updates historical information if needed.
// TODO: Update flags.
const ProbabilityEntry createUpdatedEntryFrom(
const ProbabilityEntry *const originalProbabilityEntry,
const ProbabilityEntry *const probabilityEntry) const;
bool updatePtNodeFlags(const int ptNodePos, const bool isTerminal, const bool hasMultipleChars);
static const int CHILDREN_POSITION_FIELD_SIZE;
BufferWithExtendableBuffer *const mTrieBuffer;
Ver4DictBuffers *const mBuffers;
const HeaderPolicy *const mHeaderPolicy;
DynamicPtReadingHelper mReadingHelper;
Ver4ShortcutListPolicy *const mShortcutPolicy;
};

View File

@ -43,7 +43,6 @@ const char *const Ver4PatriciaTriePolicy::MAX_BIGRAM_COUNT_QUERY = "MAX_BIGRAM_C
const int Ver4PatriciaTriePolicy::MARGIN_TO_REFUSE_DYNAMIC_OPERATIONS = 1024;
const int Ver4PatriciaTriePolicy::MIN_DICT_SIZE_TO_REFUSE_DYNAMIC_OPERATIONS =
Ver4DictConstants::MAX_DICTIONARY_SIZE - MARGIN_TO_REFUSE_DYNAMIC_OPERATIONS;
const int Ver4PatriciaTriePolicy::DUMMY_PROBABILITY_FOR_VALID_WORDS = 1;
void Ver4PatriciaTriePolicy::createAndGetAllChildDicNodes(const DicNode *const dicNode,
DicNodeVector *const childDicNodes) const {
@ -151,8 +150,7 @@ void Ver4PatriciaTriePolicy::iterateNgramEntries(const WordIdArrayView prevWordI
}
const int probability = probabilityEntry.hasHistoricalInfo() ?
ForgettingCurveUtils::decodeProbability(
probabilityEntry.getHistoricalInfo(), mHeaderPolicy)
+ ForgettingCurveUtils::getProbabilityBiasForNgram(i + 1 /* n */) :
probabilityEntry.getHistoricalInfo(), mHeaderPolicy) :
probabilityEntry.getProbability();
listener->onVisitEntry(probability, entry.getWordId());
}
@ -371,25 +369,44 @@ bool Ver4PatriciaTriePolicy::updateEntriesForWordWithNgramContext(
"dictionary.");
return false;
}
// TODO: Have count up method in language model dict content.
const int probability = isValidWord ? DUMMY_PROBABILITY_FOR_VALID_WORDS : NOT_A_PROBABILITY;
const bool updateAsAValidWord = ngramContext->isNthPrevWordBeginningOfSentence(1 /* n */) ?
false : isValidWord;
int wordId = getWordId(wordCodePoints, false /* tryLowerCaseSearch */);
if (wordId == NOT_A_WORD_ID) {
// The word is not in the dictionary.
const UnigramProperty unigramProperty(false /* representsBeginningOfSentence */,
false /* isNotAWord */, false /* isBlacklisted */, probability, historicalInfo);
false /* isNotAWord */, false /* isBlacklisted */, NOT_A_PROBABILITY,
HistoricalInfo(historicalInfo.getTimestamp(), 0 /* level */, 0 /* count */));
if (!addUnigramEntry(wordCodePoints, &unigramProperty)) {
AKLOGE("Cannot update unigarm entry in updateEntriesForWordWithNgramContext().");
AKLOGE("Cannot add unigarm entry in updateEntriesForWordWithNgramContext().");
return false;
}
const int probabilityForNgram = ngramContext->isNthPrevWordBeginningOfSentence(1 /* n */)
? NOT_A_PROBABILITY : probability;
const NgramProperty ngramProperty(wordCodePoints.toVector(), probabilityForNgram,
historicalInfo);
for (size_t i = 1; i <= ngramContext->getPrevWordCount(); ++i) {
const NgramContext trimmedNgramContext(ngramContext->getTrimmedNgramContext(i));
if (!addNgramEntry(&trimmedNgramContext, &ngramProperty)) {
AKLOGE("Cannot update ngram entry in updateEntriesForWordWithNgramContext().");
wordId = getWordId(wordCodePoints, false /* tryLowerCaseSearch */);
}
WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> prevWordIdArray;
const WordIdArrayView prevWordIds = ngramContext->getPrevWordIds(this, &prevWordIdArray,
false /* tryLowerCaseSearch */);
if (prevWordIds.firstOrDefault(NOT_A_WORD_ID) == NOT_A_WORD_ID
&& ngramContext->isNthPrevWordBeginningOfSentence(1 /* n */)) {
const UnigramProperty beginningOfSentenceUnigramProperty(
true /* representsBeginningOfSentence */,
true /* isNotAWord */, false /* isBlacklisted */, NOT_A_PROBABILITY,
HistoricalInfo(historicalInfo.getTimestamp(), 0 /* level */, 0 /* count */));
if (!addUnigramEntry(ngramContext->getNthPrevWordCodePoints(1 /* n */),
&beginningOfSentenceUnigramProperty)) {
AKLOGE("Cannot add BoS entry in updateEntriesForWordWithNgramContext().");
return false;
}
// Refresh word ids.
ngramContext->getPrevWordIds(this, &prevWordIdArray, false /* tryLowerCaseSearch */);
}
int addedNewNgramEntryCount = 0;
if (!mBuffers->getMutableLanguageModelDictContent()->updateAllEntriesOnInputWord(prevWordIds,
wordId, updateAsAValidWord, historicalInfo, mHeaderPolicy, &addedNewNgramEntryCount)) {
return false;
}
mBigramCount += addedNewNgramEntryCount;
return true;
}

View File

@ -47,8 +47,8 @@ class Ver4PatriciaTriePolicy : public DictionaryStructureWithBufferPolicy {
mShortcutPolicy(mBuffers->getMutableShortcutDictContent(),
mBuffers->getTerminalPositionLookupTable()),
mNodeReader(mDictBuffer), mPtNodeArrayReader(mDictBuffer),
mNodeWriter(mDictBuffer, mBuffers.get(), mHeaderPolicy, &mNodeReader,
&mPtNodeArrayReader, &mShortcutPolicy),
mNodeWriter(mDictBuffer, mBuffers.get(), &mNodeReader, &mPtNodeArrayReader,
&mShortcutPolicy),
mUpdatingHelper(mDictBuffer, &mNodeReader, &mNodeWriter),
mWritingHelper(mBuffers.get()),
mUnigramCount(mHeaderPolicy->getUnigramCount()),
@ -131,8 +131,6 @@ class Ver4PatriciaTriePolicy : public DictionaryStructureWithBufferPolicy {
// prevent the dictionary from overflowing.
static const int MARGIN_TO_REFUSE_DYNAMIC_OPERATIONS;
static const int MIN_DICT_SIZE_TO_REFUSE_DYNAMIC_OPERATIONS;
// TODO: Remove
static const int DUMMY_PROBABILITY_FOR_VALID_WORDS;
const Ver4DictBuffers::Ver4DictBuffersPtr mBuffers;
const HeaderPolicy *const mHeaderPolicy;
@ -144,6 +142,7 @@ class Ver4PatriciaTriePolicy : public DictionaryStructureWithBufferPolicy {
DynamicPtUpdatingHelper mUpdatingHelper;
Ver4PatriciaTrieWritingHelper mWritingHelper;
int mUnigramCount;
// TODO: Support counting ngram entries.
int mBigramCount;
std::vector<int> mTerminalPtNodePositionsForIteratingWords;
mutable bool mIsCorrupted;

View File

@ -78,11 +78,11 @@ bool Ver4PatriciaTrieWritingHelper::runGC(const int rootPtNodeArrayPos,
Ver4ShortcutListPolicy shortcutPolicy(mBuffers->getMutableShortcutDictContent(),
mBuffers->getTerminalPositionLookupTable());
Ver4PatriciaTrieNodeWriter ptNodeWriter(mBuffers->getWritableTrieBuffer(),
mBuffers, headerPolicy, &ptNodeReader, &ptNodeArrayReader, &shortcutPolicy);
mBuffers, &ptNodeReader, &ptNodeArrayReader, &shortcutPolicy);
int entryCountTable[MAX_PREV_WORD_COUNT_FOR_N_GRAM + 1];
if (!mBuffers->getMutableLanguageModelDictContent()->updateAllProbabilityEntries(headerPolicy,
entryCountTable)) {
if (!mBuffers->getMutableLanguageModelDictContent()->updateAllProbabilityEntriesForGC(
headerPolicy, entryCountTable)) {
AKLOGE("Failed to update probabilities in language model dict content.");
return false;
}
@ -118,7 +118,7 @@ bool Ver4PatriciaTrieWritingHelper::runGC(const int rootPtNodeArrayPos,
PtNodeWriter::DictPositionRelocationMap dictPositionRelocationMap;
readingHelper.initWithPtNodeArrayPos(rootPtNodeArrayPos);
Ver4PatriciaTrieNodeWriter ptNodeWriterForNewBuffers(buffersToWrite->getWritableTrieBuffer(),
buffersToWrite, headerPolicy, &ptNodeReader, &ptNodeArrayReader, &shortcutPolicy);
buffersToWrite, &ptNodeReader, &ptNodeArrayReader, &shortcutPolicy);
DynamicPtGcEventListeners::TraversePolicyToPlaceAndWriteValidPtNodesToBuffer
traversePolicyToPlaceAndWriteValidPtNodesToBuffer(&ptNodeWriterForNewBuffers,
buffersToWrite->getWritableTrieBuffer(), &dictPositionRelocationMap);
@ -133,7 +133,7 @@ bool Ver4PatriciaTrieWritingHelper::runGC(const int rootPtNodeArrayPos,
Ver4ShortcutListPolicy newShortcutPolicy(buffersToWrite->getMutableShortcutDictContent(),
buffersToWrite->getTerminalPositionLookupTable());
Ver4PatriciaTrieNodeWriter newPtNodeWriter(buffersToWrite->getWritableTrieBuffer(),
buffersToWrite, headerPolicy, &newPtNodeReader, &newPtNodeArrayreader,
buffersToWrite, &newPtNodeReader, &newPtNodeArrayreader,
&newShortcutPolicy);
// Re-assign terminal IDs for valid terminal PtNodes.
TerminalPositionLookupTable::TerminalIdMap terminalIdMap;