diff --git a/native/jni/src/proximity_info_params.cpp b/native/jni/src/proximity_info_params.cpp index 1410ab575..f9a4352ee 100644 --- a/native/jni/src/proximity_info_params.cpp +++ b/native/jni/src/proximity_info_params.cpp @@ -88,7 +88,7 @@ const float ProximityInfoParams::SKIP_PROBABALITY_WEIGHT_FOR_PROBABILITY_GAIN = // Used by ProximityInfoStateUtils::getMostProbableString() const float ProximityInfoParams::DEMOTION_LOG_PROBABILITY = 0.3f; -// Used by ProximityInfoStateUtils::updateSampledSearchKeysVector() +// Used by ProximityInfoStateUtils::updateSampledSearchKeySets() // TODO: Investigate if this is required const float ProximityInfoParams::SEARCH_KEY_RADIUS_RATIO = 0.95f; diff --git a/native/jni/src/proximity_info_params.h b/native/jni/src/proximity_info_params.h index 7c26208a8..e7aec0976 100644 --- a/native/jni/src/proximity_info_params.h +++ b/native/jni/src/proximity_info_params.h @@ -90,7 +90,7 @@ class ProximityInfoParams { // Used by ProximityInfoStateUtils::getMostProbableString() static const float DEMOTION_LOG_PROBABILITY; - // Used by ProximityInfoStateUtils::updateSampledSearchKeysVector() + // Used by ProximityInfoStateUtils::updateSampledSearchKeySets() static const float SEARCH_KEY_RADIUS_RATIO; // Used by ProximityInfoStateUtils::calculateBeelineSpeedRate() diff --git a/native/jni/src/proximity_info_state.cpp b/native/jni/src/proximity_info_state.cpp index 4c1ffb30e..bdbf8b170 100644 --- a/native/jni/src/proximity_info_state.cpp +++ b/native/jni/src/proximity_info_state.cpp @@ -16,6 +16,7 @@ #include // for memset() and memcpy() #include // for debug prints +#include #define LOG_TAG "LatinIME: proximity_info_state.cpp" @@ -75,8 +76,8 @@ void ProximityInfoState::initInputParams(const int pointerId, const float maxPoi mSampledInputIndice.clear(); mSampledLengthCache.clear(); mSampledDistanceCache_G.clear(); - mSampledNearKeysVector.clear(); - mSampledSearchKeysVector.clear(); + mSampledNearKeySets.clear(); + mSampledSearchKeySets.clear(); mSpeedRates.clear(); mBeelineSpeedPercentiles.clear(); mCharProbabilities.clear(); @@ -109,7 +110,7 @@ void ProximityInfoState::initInputParams(const int pointerId, const float maxPoi if (mSampledInputSize > 0) { ProximityInfoStateUtils::initGeometricDistanceInfos(mProximityInfo, mSampledInputSize, - lastSavedInputSize, &mSampledInputXs, &mSampledInputYs, &mSampledNearKeysVector, + lastSavedInputSize, &mSampledInputXs, &mSampledInputYs, &mSampledNearKeySets, &mSampledDistanceCache_G); if (isGeometric) { // updates probabilities of skipping or mapping each key for all points. @@ -117,10 +118,11 @@ void ProximityInfoState::initInputParams(const int pointerId, const float maxPoi mMaxPointToKeyLength, mProximityInfo->getMostCommonKeyWidth(), mProximityInfo->getKeyCount(), lastSavedInputSize, mSampledInputSize, &mSampledInputXs, &mSampledInputYs, &mSpeedRates, &mSampledLengthCache, - &mSampledDistanceCache_G, &mSampledNearKeysVector, &mCharProbabilities); - ProximityInfoStateUtils::updateSampledSearchKeysVector(mProximityInfo, + &mSampledDistanceCache_G, &mSampledNearKeySets, &mCharProbabilities); + ProximityInfoStateUtils::updateSampledSearchKeySets(mProximityInfo, mSampledInputSize, lastSavedInputSize, &mSampledLengthCache, - &mSampledNearKeysVector, &mSampledSearchKeysVector); + &mSampledNearKeySets, &mSampledSearchKeySets, + &mSampledSearchKeyVectors); mMostProbableStringProbability = ProximityInfoStateUtils::getMostProbableString( mProximityInfo, mSampledInputSize, &mCharProbabilities, mMostProbableString); @@ -245,36 +247,9 @@ ProximityType ProximityInfoState::getMatchedProximityId(const int index, const i return UNRELATED_CHAR; } -// Puts possible characters into filter and returns new filter size. -int ProximityInfoState::getAllPossibleChars( - const size_t index, int *const filter, const int filterSize) const { - if (index >= mSampledInputXs.size()) { - return filterSize; - } - int newFilterSize = filterSize; - const int keyCount = mProximityInfo->getKeyCount(); - for (int j = 0; j < keyCount; ++j) { - if (mSampledSearchKeysVector[index].test(j)) { - const int keyCodePoint = mProximityInfo->getCodePointOf(j); - bool insert = true; - // TODO: Avoid linear search - for (int k = 0; k < filterSize; ++k) { - if (filter[k] == keyCodePoint) { - insert = false; - break; - } - } - if (insert) { - filter[newFilterSize++] = keyCodePoint; - } - } - } - return newFilterSize; -} - bool ProximityInfoState::isKeyInSerchKeysAfterIndex(const int index, const int keyId) const { ASSERT(keyId >= 0 && index >= 0 && index < mSampledInputSize); - return mSampledSearchKeysVector[index].test(keyId); + return mSampledSearchKeySets[index].test(keyId); } float ProximityInfoState::getDirection(const int index0, const int index1) const { diff --git a/native/jni/src/proximity_info_state.h b/native/jni/src/proximity_info_state.h index 0386450bd..c4cbd582d 100644 --- a/native/jni/src/proximity_info_state.h +++ b/native/jni/src/proximity_info_state.h @@ -50,7 +50,7 @@ class ProximityInfoState { mIsContinuationPossible(false), mSampledInputXs(), mSampledInputYs(), mSampledTimes(), mSampledInputIndice(), mSampledLengthCache(), mBeelineSpeedPercentiles(), mSampledDistanceCache_G(), mSpeedRates(), mDirections(), mCharProbabilities(), - mSampledNearKeysVector(), mSampledSearchKeysVector(), + mSampledNearKeySets(), mSampledSearchKeySets(), mSampledSearchKeyVectors(), mTouchPositionCorrectionEnabled(false), mSampledInputSize(0), mMostProbableStringProbability(0.0f) { memset(mInputProximities, 0, sizeof(mInputProximities)); @@ -155,7 +155,9 @@ class ProximityInfoState { ProximityType getMatchedProximityId(const int index, const int c, const bool checkProximityChars, int *proximityIndex = 0) const; - int getAllPossibleChars(const size_t startIndex, int *const filter, const int filterSize) const; + const std::vector *getSearchKeyVector(const int index) const { + return &mSampledSearchKeyVectors[index]; + } float getSpeedRate(const int index) const { return mSpeedRates[index]; @@ -236,13 +238,14 @@ class ProximityInfoState { std::vector > mCharProbabilities; // The vector for the key code set which holds nearby keys for each sampled input point // 1. Used to calculate the probability of the key - // 2. Used to calculate mSampledSearchKeysVector - std::vector mSampledNearKeysVector; + // 2. Used to calculate mSampledSearchKeySets + std::vector mSampledNearKeySets; // The vector for the key code set which holds nearby keys of some trailing sampled input points // for each sampled input point. These nearby keys contain the next characters which can be in // the dictionary. Specifically, currently we are looking for keys nearby trailing sampled // inputs including the current input point. - std::vector mSampledSearchKeysVector; + std::vector mSampledSearchKeySets; + std::vector > mSampledSearchKeyVectors; bool mTouchPositionCorrectionEnabled; int mInputProximities[MAX_PROXIMITY_CHARS_SIZE * MAX_WORD_LENGTH]; int mNormalizedSquaredDistances[MAX_PROXIMITY_CHARS_SIZE * MAX_WORD_LENGTH]; diff --git a/native/jni/src/proximity_info_state_utils.cpp b/native/jni/src/proximity_info_state_utils.cpp index f9b69d264..2bf327fcc 100644 --- a/native/jni/src/proximity_info_state_utils.cpp +++ b/native/jni/src/proximity_info_state_utils.cpp @@ -224,13 +224,13 @@ namespace latinime { const ProximityInfo *const proximityInfo, const int sampledInputSize, const int lastSavedInputSize, const std::vector *const sampledInputXs, const std::vector *const sampledInputYs, - std::vector *SampledNearKeysVector, + std::vector *SampledNearKeySets, std::vector *SampledDistanceCache_G) { - SampledNearKeysVector->resize(sampledInputSize); + SampledNearKeySets->resize(sampledInputSize); const int keyCount = proximityInfo->getKeyCount(); SampledDistanceCache_G->resize(sampledInputSize * keyCount); for (int i = lastSavedInputSize; i < sampledInputSize; ++i) { - (*SampledNearKeysVector)[i].reset(); + (*SampledNearKeySets)[i].reset(); for (int k = 0; k < keyCount; ++k) { const int index = i * keyCount + k; const int x = (*sampledInputXs)[i]; @@ -240,7 +240,7 @@ namespace latinime { (*SampledDistanceCache_G)[index] = normalizedSquaredDistance; if (normalizedSquaredDistance < ProximityInfoParams::NEAR_KEY_NORMALIZED_SQUARED_THRESHOLD) { - (*SampledNearKeysVector)[i][k] = true; + (*SampledNearKeySets)[i][k] = true; } } } @@ -664,7 +664,7 @@ namespace latinime { const std::vector *const sampledSpeedRates, const std::vector *const sampledLengthCache, const std::vector *const SampledDistanceCache_G, - std::vector *SampledNearKeysVector, + std::vector *SampledNearKeySets, std::vector > *charProbabilities) { charProbabilities->resize(sampledInputSize); // Calculates probabilities of using a point as a correlated point with the character @@ -680,7 +680,7 @@ namespace latinime { float nearestKeyDistance = static_cast(MAX_POINT_TO_KEY_LENGTH); for (int j = 0; j < keyCount; ++j) { - if ((*SampledNearKeysVector)[i].test(j)) { + if ((*SampledNearKeySets)[i].test(j)) { const float distance = getPointToKeyByIdLength( maxPointToKeyLength, SampledDistanceCache_G, keyCount, i, j); if (distance < nearestKeyDistance) { @@ -761,7 +761,7 @@ namespace latinime { // Summing up probability densities of all near keys. float sumOfProbabilityDensities = 0.0f; for (int j = 0; j < keyCount; ++j) { - if ((*SampledNearKeysVector)[i].test(j)) { + if ((*SampledNearKeySets)[i].test(j)) { float distance = sqrtf(getPointToKeyByIdLength( maxPointToKeyLength, SampledDistanceCache_G, keyCount, i, j)); if (i == 0 && i != sampledInputSize - 1) { @@ -801,7 +801,7 @@ namespace latinime { // Split the probability of an input point to keys that are close to the input point. for (int j = 0; j < keyCount; ++j) { - if ((*SampledNearKeysVector)[i].test(j)) { + if ((*SampledNearKeySets)[i].test(j)) { float distance = sqrtf(getPointToKeyByIdLength( maxPointToKeyLength, SampledDistanceCache_G, keyCount, i, j)); if (i == 0 && i != sampledInputSize - 1) { @@ -885,10 +885,10 @@ namespace latinime { for (int j = 0; j < keyCount; ++j) { hash_map_compat::iterator it = (*charProbabilities)[i].find(j); if (it == (*charProbabilities)[i].end()){ - (*SampledNearKeysVector)[i].reset(j); + (*SampledNearKeySets)[i].reset(j); } else if(it->second < ProximityInfoParams::MIN_PROBABILITY) { // Erases from near keys vector because it has very low probability. - (*SampledNearKeysVector)[i].reset(j); + (*SampledNearKeySets)[i].reset(j); (*charProbabilities)[i].erase(j); } else { it->second = -logf(it->second); @@ -898,26 +898,42 @@ namespace latinime { } } -/* static */ void ProximityInfoStateUtils::updateSampledSearchKeysVector( +/* static */ void ProximityInfoStateUtils::updateSampledSearchKeySets( const ProximityInfo *const proximityInfo, const int sampledInputSize, const int lastSavedInputSize, const std::vector *const sampledLengthCache, - const std::vector *const SampledNearKeysVector, - std::vector *sampledSearchKeysVector) { - sampledSearchKeysVector->resize(sampledInputSize); + const std::vector *const SampledNearKeySets, + std::vector *sampledSearchKeySets, + std::vector > *sampledSearchKeyVectors) { + sampledSearchKeySets->resize(sampledInputSize); + sampledSearchKeyVectors->resize(sampledInputSize); const int readForwordLength = static_cast( hypotf(proximityInfo->getKeyboardWidth(), proximityInfo->getKeyboardHeight()) * ProximityInfoParams::SEARCH_KEY_RADIUS_RATIO); for (int i = 0; i < sampledInputSize; ++i) { if (i >= lastSavedInputSize) { - (*sampledSearchKeysVector)[i].reset(); + (*sampledSearchKeySets)[i].reset(); } for (int j = max(i, lastSavedInputSize); j < sampledInputSize; ++j) { // TODO: Investigate if this is required. This may not fail. if ((*sampledLengthCache)[j] - (*sampledLengthCache)[i] >= readForwordLength) { break; } - (*sampledSearchKeysVector)[i] |= (*SampledNearKeysVector)[j]; + (*sampledSearchKeySets)[i] |= (*SampledNearKeySets)[j]; + } + } + const int keyCount = proximityInfo->getKeyCount(); + for (int i = 0; i < sampledInputSize; ++i) { + std::vector *searchKeyVector = &(*sampledSearchKeyVectors)[i]; + searchKeyVector->clear(); + for (int j = 0; j < keyCount; ++j) { + if ((*sampledSearchKeySets)[i].test(j)) { + const int keyCodePoint = proximityInfo->getCodePointOf(j); + if (std::find(searchKeyVector->begin(), searchKeyVector->end(), keyCodePoint) + == searchKeyVector->end()) { + searchKeyVector->push_back(keyCodePoint); + } + } } } } diff --git a/native/jni/src/proximity_info_state_utils.h b/native/jni/src/proximity_info_state_utils.h index af0acc788..d55730aca 100644 --- a/native/jni/src/proximity_info_state_utils.h +++ b/native/jni/src/proximity_info_state_utils.h @@ -71,13 +71,14 @@ class ProximityInfoStateUtils { const std::vector *const sampledSpeedRates, const std::vector *const sampledLengthCache, const std::vector *const SampledDistanceCache_G, - std::vector *SampledNearKeysVector, + std::vector *SampledNearKeySets, std::vector > *charProbabilities); - static void updateSampledSearchKeysVector(const ProximityInfo *const proximityInfo, + static void updateSampledSearchKeySets(const ProximityInfo *const proximityInfo, const int sampledInputSize, const int lastSavedInputSize, const std::vector *const sampledLengthCache, - const std::vector *const SampledNearKeysVector, - std::vector *sampledSearchKeysVector); + const std::vector *const SampledNearKeySets, + std::vector *sampledSearchKeySets, + std::vector > *sampledSearchKeyVectors); static float getPointToKeyByIdLength(const float maxPointToKeyLength, const std::vector *const SampledDistanceCache_G, const int keyCount, const int inputIndex, const int keyId, const float scale); @@ -88,7 +89,7 @@ class ProximityInfoStateUtils { const int sampledInputSize, const int lastSavedInputSize, const std::vector *const sampledInputXs, const std::vector *const sampledInputYs, - std::vector *SampledNearKeysVector, + std::vector *SampledNearKeySets, std::vector *SampledDistanceCache_G); static void initPrimaryInputWord(const int inputSize, const int *const inputProximities, int *primaryInputWord);