1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178

use bitflags::bitflags;
use std::collections::HashMap;

use crate::types::*;

#[derive(Clone, Debug)]
pub struct VocabValue {
    pub text: String,

    /// A version of the text normalized to the alphabet
    pub norm: NormString,

    /// The absolute frequency count
    pub frequency: u32,

    /// The number of words
    pub tokencount: u8,

    /// Bitflag to indicate which lexicons match (can refer to multiple lexicons)
    pub lexindex: u32,

    /// Pointer to other vocabulary items that are considered a variant
    /// of this one (with a certain score between 0 and 1). This structure is used when loading variant/error lists
    /// and not in normal operation.
    pub variants: Option<Vec<VariantReference>>,

    pub vocabtype: VocabType,
}

bitflags! {
    pub struct VocabType: u8 {
        /// Indexed for variant matching
        const NONE = 0b00000000;

        /// Indexed for variant matching
        const INDEXED = 0b00000001;

        /// Used for Language Modelling
        const LM = 0b00000010;

        /// Marks this entry as transparent; transparent entries will only be used to find further explicitly provided variants
        /// and will never be returned as a solution by itself. For example, all erroneous variants in
        /// an errorlist are marked as intermediate.
        const TRANSPARENT = 0b00000100;
    }
}

impl VocabType {
    pub fn check(&self, test: VocabType) -> bool {
        *self & test == test
    }
}

impl From<VocabType> for bool {
    fn from(v: VocabType) -> bool {
        v != VocabType::NONE
    }
}

impl VocabValue {
    pub fn new(text: String, vocabtype: VocabType) -> Self {
        let tokencount = text.chars().filter(|c| *c == ' ').count() as u8;
        VocabValue {
            text: text,
            norm: Vec::new(),
            frequency: 1, //smoothing
            tokencount,
            lexindex: 0,
            variants: None,
            vocabtype,
        }
    }

    pub fn in_lexicon(&self, index: u8) -> bool {
        self.lexindex & (1 << index) == 1 << index
    }

    pub fn lexindex_as_vec(&self) -> Vec<u8> {
        let mut v = Vec::new();
        for i in 0..31 {
            if self.in_lexicon(i) {
                v.push(i);
            }
        }
        v
    }
}

///Map integers (indices correspond to VocabId) to string values (and optionally a frequency count)
pub type VocabDecoder = Vec<VocabValue>;

///Maps strings to integers
pub type VocabEncoder = HashMap<String, VocabId>;

///Frequency handling in case of duplicate items (may be across multiple lexicons), the
///associated with it.
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub enum FrequencyHandling {
    Sum,
    Max,
    Min,
    Replace,
}

#[derive(Clone, Debug)]
pub struct VocabParams {
    ///Column containing the Text (if any, 0-indexed)
    pub text_column: u8,
    ///Column containing the absolute frequency (if any, 0-indexed)
    pub freq_column: Option<u8>,
    ///Frequency handling in case of duplicate items (may be across multiple lexicons)
    pub freq_handling: FrequencyHandling,
    pub vocab_type: VocabType,
    /// Lexicon index
    pub index: u8,
}

impl Default for VocabParams {
    fn default() -> Self {
        Self {
            text_column: 0,
            freq_column: Some(1),
            freq_handling: FrequencyHandling::Max,
            vocab_type: VocabType::INDEXED,
            index: 0,
        }
    }
}

impl VocabParams {
    /// Set the vocabulary type (removes any previous values)
    pub fn with_vocab_type(mut self, vocab_type: VocabType) -> Self {
        self.vocab_type = vocab_type;
        self
    }
    pub fn with_freq_handling(mut self, freq_handling: FrequencyHandling) -> Self {
        self.freq_handling = freq_handling;
        self
    }
}

pub const BOS: VocabId = 0;
pub const EOS: VocabId = 1;
pub const UNK: VocabId = 2;

/// Adds some initial special tokens, required for basic language modelling in the 'search' stage
pub(crate) fn init_vocab(decoder: &mut VocabDecoder, encoder: &mut HashMap<String, VocabId>) {
    decoder.push(VocabValue {
        text: "<bos>".to_string(),
        norm: vec![],
        frequency: 0,
        tokencount: 1,
        lexindex: 0,
        variants: None,
        vocabtype: VocabType::NONE,
    });
    decoder.push(VocabValue {
        text: "<eos>".to_string(),
        norm: vec![],
        frequency: 0,
        tokencount: 1,
        lexindex: 0,
        variants: None,
        vocabtype: VocabType::NONE,
    });
    decoder.push(VocabValue {
        text: "<unk>".to_string(),
        norm: vec![],
        frequency: 0,
        tokencount: 1,
        lexindex: 0,
        variants: None,
        vocabtype: VocabType::NONE,
    });
    encoder.insert("<bos>".to_string(), BOS);
    encoder.insert("<eos>".to_string(), EOS);
    encoder.insert("<unk>".to_string(), UNK);
}