oxits 0.1.0

Time series classification and transformation library for Rust
Documentation 
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
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230

use std::collections::HashMap;

use crate::core::config::TfIdfVariant;
use crate::transformation::boss::{Boss, BossConfig, BossFitted};
use crate::utils::tfidf::{build_vocabulary, cosine_similarity};

/// BOSSVS (BOSS in Vector Space) classifier.
///
/// Pipeline:
/// 1. Fit BOSS to learn SFA model
/// 2. Build BOSS histograms for training data
/// 3. Compute TF-IDF representation per class
/// 4. Classify by cosine similarity to class TF-IDF vectors

#[derive(Debug, Clone)]
pub struct BossvsConfig {
    pub window_size: usize,
    pub word_size: usize,
    pub n_bins: usize,
    pub variant: TfIdfVariant,
}

impl BossvsConfig {
    pub fn new(window_size: usize) -> Self {
        Self {
            window_size,
            word_size: 4,
            n_bins: 4,
            variant: TfIdfVariant::SublinearTfIdf,
        }
    }
}

#[derive(Debug, Clone)]
pub struct BossvsFitted {
    pub boss_fitted: BossFitted,
    pub vocabulary: Vec<String>,
    pub class_vectors: Vec<(String, Vec<f64>)>,
}

pub struct Bossvs;

impl Bossvs {
    /// Fit BOSSVS classifier.
    pub fn fit(config: &BossvsConfig, x: &[Vec<f64>], y: &[String]) -> BossvsFitted {
        assert!(!x.is_empty(), "Input must have at least one sample");
        assert_eq!(x.len(), y.len(), "X and y must have same length");

        let boss_config = BossConfig {
            n_bins: config.n_bins,
            ..BossConfig::new(config.window_size, config.word_size)
        };
        let (boss_fitted, histograms) = Boss::fit_with_histograms(&boss_config, x, Some(y));

        // Build vocabulary from all histograms
        let all_words: Vec<Vec<String>> = histograms
            .iter()
            .map(|h| h.keys().cloned().collect())
            .collect();
        let vocabulary = build_vocabulary(&all_words);

        // Build word-to-index map
        let word_idx: HashMap<&str, usize> = vocabulary
            .iter()
            .enumerate()
            .map(|(i, w)| (w.as_str(), i))
            .collect();

        // Get unique classes
        let mut classes: Vec<String> = y.to_vec();
        classes.sort();
        classes.dedup();

        // Compute TF-IDF vector per class
        let n_docs = histograms.len();
        let n_features = vocabulary.len();

        // Document frequency
        let mut doc_freq = vec![0usize; n_features];
        for hist in &histograms {
            for word in hist.keys() {
                if let Some(&idx) = word_idx.get(word.as_str()) {
                    doc_freq[idx] += 1;
                }
            }
        }

        // IDF
        let idf: Vec<f64> = doc_freq
            .iter()
            .map(|&df| ((1.0 + n_docs as f64) / (1.0 + df as f64)).ln() + 1.0)
            .collect();

        // Per-class aggregated TF-IDF vectors
        let mut class_vectors = Vec::new();
        for class in &classes {
            let mut class_vec = vec![0.0; n_features];

            for (i, label) in y.iter().enumerate() {
                if label == class {
                    for (word, &count) in &histograms[i] {
                        if let Some(&idx) = word_idx.get(word.as_str()) {
                            let tf = match config.variant {
                                TfIdfVariant::SublinearTfIdf => {
                                    if count > 0 {
                                        1.0 + (count as f64).ln()
                                    } else {
                                        0.0
                                    }
                                }
                                _ => count as f64,
                            };
                            class_vec[idx] += tf * idf[idx];
                        }
                    }
                }
            }

            // L2 normalize
            let norm: f64 = class_vec.iter().map(|v| v * v).sum::<f64>().sqrt();
            if norm > 0.0 {
                for v in &mut class_vec {
                    *v /= norm;
                }
            }

            class_vectors.push((class.clone(), class_vec));
        }

        BossvsFitted {
            boss_fitted,
            vocabulary,
            class_vectors,
        }
    }

    /// Predict class labels.
    pub fn predict(fitted: &BossvsFitted, x: &[Vec<f64>]) -> Vec<String> {
        let histograms = Boss::transform(&fitted.boss_fitted, x);

        let word_idx: HashMap<&str, usize> = fitted
            .vocabulary
            .iter()
            .enumerate()
            .map(|(i, w)| (w.as_str(), i))
            .collect();
        let n_features = fitted.vocabulary.len();

        // IDF from fitted vocabulary
        histograms
            .iter()
            .map(|hist| {
                // Build sample vector
                let mut sample_vec = vec![0.0; n_features];
                for (word, &count) in hist {
                    if let Some(&idx) = word_idx.get(word.as_str()) {
                        sample_vec[idx] = count as f64;
                    }
                }

                // Find most similar class
                fitted
                    .class_vectors
                    .iter()
                    .map(|(class, class_vec)| {
                        (class.as_str(), cosine_similarity(&sample_vec, class_vec))
                    })
                    .max_by(|a, b| a.1.partial_cmp(&b.1).unwrap())
                    .map(|(class, _)| class.to_string())
                    .unwrap()
            })
            .collect()
    }

    /// Compute classification accuracy.
    pub fn score(fitted: &BossvsFitted, x: &[Vec<f64>], y: &[String]) -> f64 {
        let predictions = Self::predict(fitted, x);
        let correct = predictions
            .iter()
            .zip(y.iter())
            .filter(|(p, t)| p == t)
            .count();
        correct as f64 / y.len() as f64
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_bossvs_basic() {
        let config = BossvsConfig::new(4);
        let x = vec![
            vec![0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0],
            vec![0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 8.0],
            vec![7.0, 6.0, 5.0, 4.0, 3.0, 2.0, 1.0, 0.0],
            vec![8.0, 6.0, 5.0, 4.0, 3.0, 2.0, 1.0, 0.0],
        ];
        let y = vec![
            "A".to_string(),
            "A".to_string(),
            "B".to_string(),
            "B".to_string(),
        ];
        let fitted = Bossvs::fit(&config, &x, &y);
        let predictions = Bossvs::predict(&fitted, &x);
        assert_eq!(predictions.len(), 4);
    }

    #[test]
    fn test_bossvs_score() {
        let config = BossvsConfig::new(3);
        let x = vec![
            vec![0.0, 1.0, 2.0, 3.0, 4.0, 5.0],
            vec![0.0, 1.0, 2.0, 3.0, 4.0, 6.0],
            vec![5.0, 4.0, 3.0, 2.0, 1.0, 0.0],
            vec![6.0, 4.0, 3.0, 2.0, 1.0, 0.0],
        ];
        let y = vec![
            "A".to_string(),
            "A".to_string(),
            "B".to_string(),
            "B".to_string(),
        ];
        let fitted = Bossvs::fit(&config, &x, &y);
        let score = Bossvs::score(&fitted, &x, &y);
        assert!((0.0..=1.0).contains(&score));
    }
}