1 /* 2 * Copyright (C) 2017 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 LIBTEXTCLASSIFIER_SMARTSELECT_TOKEN_FEATURE_EXTRACTOR_H_ 18 #define LIBTEXTCLASSIFIER_SMARTSELECT_TOKEN_FEATURE_EXTRACTOR_H_ 19 20 #include <memory> 21 #include <vector> 22 23 #include "base.h" 24 #include "smartselect/types.h" 25 #include "util/strings/stringpiece.h" 26 #include "unicode/regex.h" 27 28 namespace libtextclassifier { 29 30 struct TokenFeatureExtractorOptions { 31 // Number of buckets used for hashing charactergrams. 32 int num_buckets = 0; 33 34 // Orders of charactergrams to extract. E.g., 2 means character bigrams, 3 35 // character trigrams etc. 36 std::vector<int> chargram_orders; 37 38 // Whether to extract the token case feature. 39 bool extract_case_feature = false; 40 41 // If true, will use the unicode-aware functionality for extracting features. 42 bool unicode_aware_features = false; 43 44 // Whether to extract the selection mask feature. 45 bool extract_selection_mask_feature = false; 46 47 // Regexp features to extract. 48 std::vector<std::string> regexp_features; 49 50 // Whether to remap digits to a single number. 51 bool remap_digits = false; 52 53 // Whether to lowercase all tokens. 54 bool lowercase_tokens = false; 55 56 // Maximum length of a word. 57 int max_word_length = 20; 58 }; 59 60 class TokenFeatureExtractor { 61 public: 62 explicit TokenFeatureExtractor(const TokenFeatureExtractorOptions& options); 63 64 // Extracts features from a token. 65 // - is_in_span is a bool indicator whether the token is a part of the 66 // selection span (true) or not (false). 67 // - sparse_features are indices into a sparse feature vector of size 68 // options.num_buckets which are set to 1.0 (others are implicitly 0.0). 69 // - dense_features are values of a dense feature vector of size 0-2 70 // (depending on the options) for the token 71 bool Extract(const Token& token, bool is_in_span, 72 std::vector<int>* sparse_features, 73 std::vector<float>* dense_features) const; 74 DenseFeaturesCount()75 int DenseFeaturesCount() const { 76 return options_.extract_case_feature + 77 options_.extract_selection_mask_feature + regex_patterns_.size(); 78 } 79 80 protected: 81 // Hashes given token to given number of buckets. 82 int HashToken(StringPiece token) const; 83 84 // Extracts the charactergram features from the token. 85 std::vector<int> ExtractCharactergramFeatures(const Token& token) const; 86 87 // Extracts the charactergram features from the token in a non-unicode-aware 88 // way. 89 std::vector<int> ExtractCharactergramFeaturesAscii(const Token& token) const; 90 91 // Extracts the charactergram features from the token in a unicode-aware way. 92 std::vector<int> ExtractCharactergramFeaturesUnicode( 93 const Token& token) const; 94 95 private: 96 TokenFeatureExtractorOptions options_; 97 98 std::vector<std::unique_ptr<icu::RegexPattern>> regex_patterns_; 99 }; 100 101 } // namespace libtextclassifier 102 103 #endif // LIBTEXTCLASSIFIER_SMARTSELECT_TOKEN_FEATURE_EXTRACTOR_H_ 104