fraug/augmenters/
quantize.rs

1use super::base::Augmenter;
2use tracing::{info_span};
3/// Quantize time series to a level set
4///
5/// The level set is constructed by uniformly discretizing the range of all values in the series
6pub struct Quantize {
7    pub name: String,
8    /// Number of levels in the level set
9    levels: usize,
10    p: f64,
11}
12
13impl Quantize {
14    /// Creates new quantize augmenter
15    pub fn new(levels: usize) -> Self {
16        Quantize {
17            name: "Quantize".to_string(),
18            levels,
19            p: 1.0,
20        }
21    }
22}
23
24impl Augmenter for Quantize {
25    fn augment_one(&self, x: &[f64]) -> Vec<f64> {
26        let span = info_span!("", step = "augment_one");
27        let _enter = span.enter();
28        let max = x.iter().fold(f64::NEG_INFINITY, |a, &b| a.max(b));
29        let min = x.iter().fold(f64::INFINITY, |a, &b| a.min(b));
30        let range = max - min;
31        let step = range / self.levels as f64;
32        let level_set = (0..self.levels)
33            .map(|level| min + level as f64 * step)
34            .collect::<Vec<_>>();
35
36        // Could be faster using e.g. binary search
37        x.iter()
38            .map(|v| {
39                let i = level_set
40                    .iter()
41                    .map(|&l| (l - *v).abs())
42                    .enumerate()
43                    .fold(
44                        (0, f64::INFINITY),
45                        |(i, a), (j, b)| if a > b { (j, b) } else { (i, a) },
46                    )
47                    .0;
48                level_set[i]
49            })
50            .collect::<Vec<_>>()
51    }
52
53    fn get_probability(&self) -> f64 {
54        self.p
55    }
56
57    fn set_probability(&mut self, probability: f64) {
58        self.p = probability;
59    }
60
61    fn get_name(&self) ->String {
62        self.name.clone()
63    }
64}
fraug/augmenters/quantize.rs

fraug/augmenters/
quantize.rs
