aboutsummaryrefslogtreecommitdiffstats
path: root/native/jni/src
diff options
context:
space:
mode:
Diffstat (limited to 'native/jni/src')
-rw-r--r--native/jni/src/bigram_dictionary.cpp14
-rw-r--r--native/jni/src/binary_format.h27
-rw-r--r--native/jni/src/dictionary.cpp4
-rw-r--r--native/jni/src/dictionary.h2
-rw-r--r--native/jni/src/unigram_dictionary.cpp17
-rw-r--r--native/jni/src/unigram_dictionary.h2
6 files changed, 47 insertions, 19 deletions
diff --git a/native/jni/src/bigram_dictionary.cpp b/native/jni/src/bigram_dictionary.cpp
index ac2a26172..eb4bf8d1a 100644
--- a/native/jni/src/bigram_dictionary.cpp
+++ b/native/jni/src/bigram_dictionary.cpp
@@ -117,14 +117,22 @@ int BigramDictionary::getBigrams(const int32_t *prevWord, int prevWordLength, in
do {
bigramFlags = BinaryFormat::getFlagsAndForwardPointer(root, &pos);
uint16_t bigramBuffer[MAX_WORD_LENGTH];
+ int unigramFreq;
const int bigramPos = BinaryFormat::getAttributeAddressAndForwardPointer(root, bigramFlags,
&pos);
const int length = BinaryFormat::getWordAtAddress(root, bigramPos, MAX_WORD_LENGTH,
- bigramBuffer);
+ bigramBuffer, &unigramFreq);
// codesSize == 0 means we are trying to find bigram predictions.
if (codesSize < 1 || checkFirstCharacter(bigramBuffer)) {
- const int frequency = UnigramDictionary::MASK_ATTRIBUTE_FREQUENCY & bigramFlags;
+ const int bigramFreq = UnigramDictionary::MASK_ATTRIBUTE_FREQUENCY & bigramFlags;
+ // Due to space constraints, the frequency for bigrams is approximate - the lower the
+ // unigram frequency, the worse the precision. The theoritical maximum error in
+ // resulting frequency is 8 - although in the practice it's never bigger than 3 or 4
+ // in very bad cases. This means that sometimes, we'll see some bigrams interverted
+ // here, but it can't get too bad.
+ const int frequency =
+ BinaryFormat::computeFrequencyForBigram(unigramFreq, bigramFreq);
if (addWordBigram(bigramBuffer, length, frequency)) {
++bigramCount;
}
@@ -149,8 +157,8 @@ int BigramDictionary::getBigramListPositionForWord(const int32_t *prevWord,
} else {
pos = BinaryFormat::skipOtherCharacters(root, pos);
}
- pos = BinaryFormat::skipChildrenPosition(flags, pos);
pos = BinaryFormat::skipFrequency(flags, pos);
+ pos = BinaryFormat::skipChildrenPosition(flags, pos);
pos = BinaryFormat::skipShortcuts(root, flags, pos);
return pos;
}
diff --git a/native/jni/src/binary_format.h b/native/jni/src/binary_format.h
index 40f197619..51bf8ebbc 100644
--- a/native/jni/src/binary_format.h
+++ b/native/jni/src/binary_format.h
@@ -66,7 +66,8 @@ class BinaryFormat {
static int getTerminalPosition(const uint8_t* const root, const int32_t* const inWord,
const int length);
static int getWordAtAddress(const uint8_t* const root, const int address, const int maxDepth,
- uint16_t* outWord);
+ uint16_t* outWord, int* outUnigramFrequency);
+ static int computeFrequencyForBigram(const int unigramFreq, const int bigramFreq);
static int getProbability(const int position, const std::map<int, int> *bigramMap,
const uint8_t *bigramFilter, const int unigramFreq);
@@ -390,10 +391,11 @@ inline int BinaryFormat::getTerminalPosition(const uint8_t* const root,
* address: the byte position of the last chargroup of the word we are searching for (this is
* what is stored as the "bigram address" in each bigram)
* outword: an array to write the found word, with MAX_WORD_LENGTH size.
+ * outUnigramFrequency: a pointer to an int to write the frequency into.
* Return value : the length of the word, of 0 if the word was not found.
*/
inline int BinaryFormat::getWordAtAddress(const uint8_t* const root, const int address,
- const int maxDepth, uint16_t* outWord) {
+ const int maxDepth, uint16_t* outWord, int* outUnigramFrequency) {
int pos = 0;
int wordPos = 0;
@@ -421,11 +423,12 @@ inline int BinaryFormat::getWordAtAddress(const uint8_t* const root, const int a
// We count chars in order to avoid infinite loops if the file is broken or
// if there is some other bug
int charCount = maxDepth;
- while (-1 != nextChar && --charCount > 0) {
+ while (NOT_A_CHARACTER != nextChar && --charCount > 0) {
outWord[++wordPos] = nextChar;
nextChar = getCharCodeAndForwardPointer(root, &pos);
}
}
+ *outUnigramFrequency = readFrequencyWithoutMovingPointer(root, pos);
return ++wordPos;
}
// We need to skip past this char group, so skip any remaining chars after the
@@ -529,6 +532,16 @@ static inline int backoff(const int unigramFreq) {
// return unigramFreq > 8 ? unigramFreq - 8 : (0 == unigramFreq ? 0 : 8);
}
+inline int BinaryFormat::computeFrequencyForBigram(const int unigramFreq, const int bigramFreq) {
+ // We divide the range [unigramFreq..255] in 16.5 steps - in other words, we want the
+ // unigram frequency to be the median value of the 17th step from the top. A value of
+ // 0 for the bigram frequency represents the middle of the 16th step from the top,
+ // while a value of 15 represents the middle of the top step.
+ // See makedict.BinaryDictInputOutput for details.
+ const float stepSize = ((float)MAX_FREQ - unigramFreq) / (1.5f + MAX_BIGRAM_FREQ);
+ return (int)(unigramFreq + (bigramFreq + 1) * stepSize);
+}
+
// This returns a probability in log space.
inline int BinaryFormat::getProbability(const int position, const std::map<int, int> *bigramMap,
const uint8_t *bigramFilter, const int unigramFreq) {
@@ -537,13 +550,7 @@ inline int BinaryFormat::getProbability(const int position, const std::map<int,
const std::map<int, int>::const_iterator bigramFreqIt = bigramMap->find(position);
if (bigramFreqIt != bigramMap->end()) {
const int bigramFreq = bigramFreqIt->second;
- // We divide the range [unigramFreq..255] in 16.5 steps - in other words, we want the
- // unigram frequency to be the median value of the 17th step from the top. A value of
- // 0 for the bigram frequency represents the middle of the 16th step from the top,
- // while a value of 15 represents the middle of the top step.
- // See makedict.BinaryDictInputOutput for details.
- const float stepSize = ((float)MAX_FREQ - unigramFreq) / (1.5f + MAX_BIGRAM_FREQ);
- return (int)(unigramFreq + bigramFreq * stepSize);
+ return computeFrequencyForBigram(unigramFreq, bigramFreq);
} else {
return backoff(unigramFreq);
}
diff --git a/native/jni/src/dictionary.cpp b/native/jni/src/dictionary.cpp
index 65d0f73a3..1fb02478b 100644
--- a/native/jni/src/dictionary.cpp
+++ b/native/jni/src/dictionary.cpp
@@ -55,8 +55,8 @@ Dictionary::~Dictionary() {
delete mBigramDictionary;
}
-bool Dictionary::isValidWord(const int32_t *word, int length) {
- return mUnigramDictionary->isValidWord(word, length);
+int Dictionary::getFrequency(const int32_t *word, int length) {
+ return mUnigramDictionary->getFrequency(word, length);
}
bool Dictionary::isValidBigram(const int32_t *word1, int length1, const int32_t *word2,
diff --git a/native/jni/src/dictionary.h b/native/jni/src/dictionary.h
index 87891ee4d..9f2367904 100644
--- a/native/jni/src/dictionary.h
+++ b/native/jni/src/dictionary.h
@@ -52,7 +52,7 @@ class Dictionary {
maxWordLength, maxBigrams);
}
- bool isValidWord(const int32_t *word, int length);
+ int getFrequency(const int32_t *word, int length);
bool isValidBigram(const int32_t *word1, int length1, const int32_t *word2, int length2);
void *getDict() { return (void *)mDict; }
int getDictSize() { return mDictSize; }
diff --git a/native/jni/src/unigram_dictionary.cpp b/native/jni/src/unigram_dictionary.cpp
index 828582848..efe9c4fe3 100644
--- a/native/jni/src/unigram_dictionary.cpp
+++ b/native/jni/src/unigram_dictionary.cpp
@@ -747,8 +747,21 @@ int UnigramDictionary::getMostFrequentWordLikeInner(const uint16_t * const inWor
return maxFreq;
}
-bool UnigramDictionary::isValidWord(const int32_t* const inWord, const int length) const {
- return NOT_VALID_WORD != BinaryFormat::getTerminalPosition(DICT_ROOT, inWord, length);
+int UnigramDictionary::getFrequency(const int32_t* const inWord, const int length) const {
+ const uint8_t* const root = DICT_ROOT;
+ int pos = BinaryFormat::getTerminalPosition(root, inWord, length);
+ if (NOT_VALID_WORD == pos) {
+ return NOT_A_PROBABILITY;
+ }
+ const uint8_t flags = BinaryFormat::getFlagsAndForwardPointer(root, &pos);
+ const bool hasMultipleChars = (0 != (FLAG_HAS_MULTIPLE_CHARS & flags));
+ if (hasMultipleChars) {
+ pos = BinaryFormat::skipOtherCharacters(root, pos);
+ } else {
+ BinaryFormat::getCharCodeAndForwardPointer(DICT_ROOT, &pos);
+ }
+ const int unigramFreq = BinaryFormat::readFrequencyWithoutMovingPointer(root, pos);
+ return unigramFreq;
}
// TODO: remove this function.
diff --git a/native/jni/src/unigram_dictionary.h b/native/jni/src/unigram_dictionary.h
index b9233518f..b70894004 100644
--- a/native/jni/src/unigram_dictionary.h
+++ b/native/jni/src/unigram_dictionary.h
@@ -72,7 +72,7 @@ class UnigramDictionary {
UnigramDictionary(const uint8_t* const streamStart, int typedLetterMultipler,
int fullWordMultiplier, int maxWordLength, int maxWords, const unsigned int flags);
- bool isValidWord(const int32_t* const inWord, const int length) const;
+ int getFrequency(const int32_t* const inWord, const int length) const;
int getBigramPosition(int pos, unsigned short *word, int offset, int length) const;
int getSuggestions(ProximityInfo *proximityInfo, WordsPriorityQueuePool *queuePool,
Correction *correction, const int *xcoordinates, const int *ycoordinates,
generated by cgit v1.2.3-83-g751a (git 2.47.1) at 2025-05-09 19:12:13 +0000