1 /*
2  * Copyright (C) 2014, 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 #ifndef LATINIME_DYNAMIC_LANGUAGE_MODEL_PROBABILITY_UTILS_H
18 #define LATINIME_DYNAMIC_LANGUAGE_MODEL_PROBABILITY_UTILS_H
19 
20 #include <algorithm>
21 
22 #include "defines.h"
23 #include "dictionary/property/historical_info.h"
24 #include "utils/ngram_utils.h"
25 #include "utils/time_keeper.h"
26 
27 namespace latinime {
28 
29 class DynamicLanguageModelProbabilityUtils {
30  public:
computeRawProbabilityFromCounts(const int count,const int contextCount,const NgramType ngramType)31     static float computeRawProbabilityFromCounts(const int count, const int contextCount,
32             const NgramType ngramType) {
33         const int minCount = ASSUMED_MIN_COUNTS[static_cast<int>(ngramType)];
34         return static_cast<float>(count) / static_cast<float>(std::max(contextCount, minCount));
35     }
36 
backoff(const int ngramProbability,const NgramType ngramType)37     static float backoff(const int ngramProbability, const NgramType ngramType) {
38         const int probability =
39                 ngramProbability + ENCODED_BACKOFF_WEIGHTS[static_cast<int>(ngramType)];
40         return std::min(std::max(probability, NOT_A_PROBABILITY), MAX_PROBABILITY);
41     }
42 
getDecayedProbability(const int probability,const HistoricalInfo historicalInfo)43     static int getDecayedProbability(const int probability, const HistoricalInfo historicalInfo) {
44         const int elapsedTime = TimeKeeper::peekCurrentTime() - historicalInfo.getTimestamp();
45         if (elapsedTime < 0) {
46             AKLOGE("The elapsed time is negatime value. Timestamp overflow?");
47             return NOT_A_PROBABILITY;
48         }
49         // TODO: Improve this logic.
50         // We don't modify probability depending on the elapsed time.
51         return probability;
52     }
53 
shouldRemoveEntryDuringGC(const HistoricalInfo historicalInfo)54     static int shouldRemoveEntryDuringGC(const HistoricalInfo historicalInfo) {
55         // TODO: Improve this logic.
56         const int elapsedTime = TimeKeeper::peekCurrentTime() - historicalInfo.getTimestamp();
57         return elapsedTime > DURATION_TO_DISCARD_ENTRY_IN_SECONDS;
58     }
59 
getPriorityToPreventFromEviction(const HistoricalInfo historicalInfo)60     static int getPriorityToPreventFromEviction(const HistoricalInfo historicalInfo) {
61         // TODO: Improve this logic.
62         // More recently input entries get higher priority.
63         return historicalInfo.getTimestamp();
64     }
65 
66 private:
67     DISALLOW_IMPLICIT_CONSTRUCTORS(DynamicLanguageModelProbabilityUtils);
68 
69     static_assert(MAX_PREV_WORD_COUNT_FOR_N_GRAM <= 3, "Max supported Ngram is Quadgram.");
70 
71     static const int ASSUMED_MIN_COUNTS[];
72     static const int ENCODED_BACKOFF_WEIGHTS[];
73     static const int DURATION_TO_DISCARD_ENTRY_IN_SECONDS;
74 };
75 
76 } // namespace latinime
77 #endif /* LATINIME_DYNAMIC_LANGUAGE_MODEL_PROBABILITY_UTILS_H */
78