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
use lazy_static::lazy_static;
use std::collections::HashMap;
use std::{borrow::Cow, cmp::min};
lazy_static! {
pub static ref DEFAULT_MODEL: LanguageModel<'static> =
LanguageModel::new_model(include_str!("../wordninja_words.txt"));
}
#[derive(Debug, Clone)]
pub struct LanguageModel<'s> {
wordcost: HashMap<&'s str, f64>,
maxword: usize,
}
impl<'s> LanguageModel<'s> {
pub fn new(words: &[&'s str]) -> Self {
let count = words.len();
let count_ln = (count as f64).ln();
let mut wordcost = HashMap::with_capacity(count);
let mut maxword = 0;
for (i, word) in words.iter().enumerate() {
let cost = (((i + 1) as f64) * count_ln).ln();
wordcost.insert(*word, cost);
if word.len() > maxword {
maxword = word.len();
}
}
Self { wordcost, maxword }
}
pub fn new_model(model: &'s str) -> Self {
let words: Vec<&str> = model.lines().filter(|s| !s.is_empty()).collect();
Self::new(&words)
}
pub fn split<'a>(&self, s: &'a str) -> Vec<Cow<'a, str>> {
s.split(|c: char| !(c.is_ascii_alphanumeric() || c == '\''))
.filter(|s| !s.is_empty())
.flat_map(|s| self.split_part(s))
.collect()
}
fn split_part<'a>(&self, s: &'a str) -> impl Iterator<Item = Cow<'a, str>> {
let mut cost = Vec::with_capacity(s.len() + 1);
cost.push(0.0);
for i in 1..s.len() + 1 {
let (c, _) = self.best_match(s, &cost, i);
cost.push(c);
}
let mut out: Vec<Cow<str>> = vec![];
let mut i = s.len();
while i > 0 {
let (c, k) = self.best_match(s, &cost, i);
assert!((c - cost[i]).abs() < f64::EPSILON);
let mut new_token = true;
let token = &s[i - k..i];
if token != "'" {
if !out.is_empty() {
let last_out = out.iter_mut().last().unwrap();
let is_s = *last_out == "'s";
let digit_follow = s.chars().nth(i - 1).unwrap().is_ascii_digit()
&& last_out.chars().next().unwrap().is_ascii_digit();
if is_s || digit_follow {
*last_out = [token, last_out].concat().into();
new_token = false;
}
}
}
if new_token {
out.push(token.into());
}
i -= k;
}
out.into_iter().rev()
}
fn best_match(&self, s: &str, cost: &[f64], i: usize) -> (f64, usize) {
let back = min(self.maxword, i);
cost[i - back..i]
.iter()
.rev()
.enumerate()
.map(|(k, c)| {
let lower = s[i - k - 1..i].to_ascii_lowercase();
let match_cost = c + self.wordcost.get(&lower[..]).unwrap_or(&f64::MAX);
(match_cost, k + 1)
})
.min_by(|(x, _), (y, _)| x.partial_cmp(y).expect("Tried to compare a NaN"))
.unwrap()
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn simple() {
assert_eq!(
DEFAULT_MODEL.split("derekanderson"),
vec!["derek", "anderson"]
);
assert_eq!(
DEFAULT_MODEL.split("DEREKANDERSON"),
vec!["DEREK", "ANDERSON"]
);
}
#[test]
fn with_underscores_etc() {
let cmp = vec!["derek", "anderson"];
assert_eq!(DEFAULT_MODEL.split("derek anderson"), cmp);
assert_eq!(DEFAULT_MODEL.split("derek-anderson"), cmp);
assert_eq!(DEFAULT_MODEL.split("derek_anderson"), cmp);
assert_eq!(DEFAULT_MODEL.split("derek/anderson"), cmp);
}
#[test]
fn digits() {
assert_eq!(
DEFAULT_MODEL.split("win32intel"),
vec!["win", "32", "intel"]
);
}
#[test]
fn apostrophes() {
assert_eq!(
DEFAULT_MODEL.split("that'sthesheriff'sbadge"),
vec!["that's", "the", "sheriff's", "badge"]
);
}
#[test]
fn custom() {
let model = LanguageModel::new_model(include_str!("../test_lang.txt"));
assert_eq!(model.split("derek"), vec!["der", "ek"]);
}
}