Home | History | Annotate | Download | only in utils 
      1 /*
      2  * Copyright (C) 2013, The Android Open Source Project
      3  *
      4  * Licensed under the Apache License, Version 2.0 (the "License");
      5  * you may not use this file except in compliance with the License.
      6  * You may obtain a copy of 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,
     12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
     13  * See the License for the specific language governing permissions and
     14  * limitations under the License.
     15  */
     16 
     17 #include "dictionary/utils/multi_bigram_map.h"
     18 
     19 #include <cstddef>
     20 #include <unordered_map>
     21 
     22 namespace latinime {
     23 
     24 // Max number of bigram maps (previous word contexts) to be cached. Increasing this number
     25 // could improve bigram lookup speed for multi-word suggestions, but at the cost of more memory
     26 // usage. Also, there are diminishing returns since the most frequently used bigrams are
     27 // typically near the beginning of the input and are thus the first ones to be cached. Note
     28 // that these bigrams are reset for each new composing word.
     29 const size_t MultiBigramMap::MAX_CACHED_PREV_WORDS_IN_BIGRAM_MAP = 25;
     30 
     31 // Most common previous word contexts currently have 100 bigrams
     32 const int MultiBigramMap::BigramMap::DEFAULT_HASH_MAP_SIZE_FOR_EACH_BIGRAM_MAP = 100;
     33 
     34 // Look up the bigram probability for the given word pair from the cached bigram maps.
     35 // Also caches the bigrams if there is space remaining and they have not been cached already.
     36 int MultiBigramMap::getBigramProbability(
     37         const DictionaryStructureWithBufferPolicy *const structurePolicy,
     38         const WordIdArrayView prevWordIds, const int nextWordId,
     39         const int unigramProbability) {
     40     if (prevWordIds.empty() || prevWordIds[0] == NOT_A_WORD_ID) {
     41         return structurePolicy->getProbability(unigramProbability, NOT_A_PROBABILITY);
     42     }
     43     const auto mapPosition = mBigramMaps.find(prevWordIds[0]);
     44     if (mapPosition != mBigramMaps.end()) {
     45         return mapPosition->second.getBigramProbability(structurePolicy, nextWordId,
     46                 unigramProbability);
     47     }
     48     if (mBigramMaps.size() < MAX_CACHED_PREV_WORDS_IN_BIGRAM_MAP) {
     49         addBigramsForWord(structurePolicy, prevWordIds);
     50         return mBigramMaps[prevWordIds[0]].getBigramProbability(structurePolicy,
     51                 nextWordId, unigramProbability);
     52     }
     53     return readBigramProbabilityFromBinaryDictionary(structurePolicy, prevWordIds,
     54             nextWordId, unigramProbability);
     55 }
     56 
     57 void MultiBigramMap::BigramMap::init(
     58         const DictionaryStructureWithBufferPolicy *const structurePolicy,
     59         const WordIdArrayView prevWordIds) {
     60     structurePolicy->iterateNgramEntries(prevWordIds, this /* listener */);
     61 }
     62 
     63 int MultiBigramMap::BigramMap::getBigramProbability(
     64         const DictionaryStructureWithBufferPolicy *const structurePolicy,
     65         const int nextWordId, const int unigramProbability) const {
     66     int bigramProbability = NOT_A_PROBABILITY;
     67     if (mBloomFilter.isInFilter(nextWordId)) {
     68         const auto bigramProbabilityIt = mBigramMap.find(nextWordId);
     69         if (bigramProbabilityIt != mBigramMap.end()) {
     70             bigramProbability = bigramProbabilityIt->second;
     71         }
     72     }
     73     return structurePolicy->getProbability(unigramProbability, bigramProbability);
     74 }
     75 
     76 void MultiBigramMap::BigramMap::onVisitEntry(const int ngramProbability, const int targetWordId) {
     77     if (targetWordId == NOT_A_WORD_ID) {
     78         return;
     79     }
     80     mBigramMap[targetWordId] = ngramProbability;
     81     mBloomFilter.setInFilter(targetWordId);
     82 }
     83 
     84 void MultiBigramMap::addBigramsForWord(
     85         const DictionaryStructureWithBufferPolicy *const structurePolicy,
     86         const WordIdArrayView prevWordIds) {
     87     mBigramMaps[prevWordIds[0]].init(structurePolicy, prevWordIds);
     88 }
     89 
     90 int MultiBigramMap::readBigramProbabilityFromBinaryDictionary(
     91         const DictionaryStructureWithBufferPolicy *const structurePolicy,
     92         const WordIdArrayView prevWordIds, const int nextWordId, const int unigramProbability) {
     93     const int bigramProbability = structurePolicy->getProbabilityOfWord(prevWordIds, nextWordId);
     94     if (bigramProbability != NOT_A_PROBABILITY) {
     95         return bigramProbability;
     96     }
     97     return structurePolicy->getProbability(unigramProbability, NOT_A_PROBABILITY);
     98 }
     99 
    100 } // namespace latinime
    101