rill-ml 0.7.1

Lightweight, serializable online machine learning for Rust applications and streaming data.
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

//! Online mean using a numerically stable incremental update.
//!
//! Time complexity per update: `O(1)`. Space complexity: `O(1)`.

use crate::error::{RillError, checked_increment, ensure_finite};
use crate::traits::OnlineStatistic;

/// Incremental mean computed with the delta method to minimize floating-point
/// accumulation error.
///
/// # Examples
///
/// ```
/// use rill_ml::stats::Mean;
/// use rill_ml::OnlineStatistic;
///
/// let mut m = Mean::new();
/// m.update(1.0).unwrap();
/// m.update(2.0).unwrap();
/// m.update(3.0).unwrap();
/// assert_eq!(m.value(), 2.0);
/// assert_eq!(m.samples_seen(), 3);
/// ```
#[derive(Debug, Clone, Default)]
#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))]
pub struct Mean {
    count: u64,
    mean: f64,
}

impl Mean {
    /// Create a new empty mean accumulator.
    pub const fn new() -> Self {
        Self {
            count: 0,
            mean: 0.0,
        }
    }

    /// Current mean, or `0.0` if no observations have been seen.
    pub const fn value(&self) -> f64 {
        self.mean
    }

    /// Number of observations seen so far.
    pub const fn count(&self) -> u64 {
        self.count
    }
}

impl OnlineStatistic for Mean {
    fn update(&mut self, value: f64) -> Result<(), RillError> {
        ensure_finite("value", value)?;
        let next_count = checked_increment(self.count, "mean sample")?;
        let delta = value - self.mean;
        ensure_finite("mean delta", delta)?;
        let next_mean = self.mean + delta / next_count as f64;
        ensure_finite("mean", next_mean)?;

        self.count = next_count;
        self.mean = next_mean;
        Ok(())
    }

    fn samples_seen(&self) -> u64 {
        self.count
    }

    fn reset(&mut self) {
        self.count = 0;
        self.mean = 0.0;
    }
}

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

    #[test]
    fn mean_of_simple_sequence() {
        let mut m = Mean::new();
        for x in [1.0, 2.0, 3.0, 4.0, 5.0] {
            m.update(x).unwrap();
        }
        assert_eq!(m.value(), 3.0);
        assert_eq!(m.count(), 5);
    }

    #[test]
    fn mean_empty_is_zero() {
        let m = Mean::new();
        assert_eq!(m.value(), 0.0);
        assert_eq!(m.count(), 0);
    }

    #[test]
    fn mean_rejects_non_finite() {
        let mut m = Mean::new();
        assert!(m.update(f64::NAN).is_err());
        assert!(m.update(f64::INFINITY).is_err());
        assert_eq!(m.count(), 0);
    }

    #[test]
    fn mean_rejects_overflow_without_mutating_state() {
        let mut m = Mean::new();
        m.update(f64::MAX).unwrap();
        let before = m.clone();
        assert!(m.update(-f64::MAX).is_err());
        assert_eq!(m.count(), before.count());
        assert_eq!(m.value(), before.value());
    }

    #[test]
    fn mean_reset() {
        let mut m = Mean::new();
        m.update(10.0).unwrap();
        m.update(20.0).unwrap();
        m.reset();
        assert_eq!(m.count(), 0);
        assert_eq!(m.value(), 0.0);
    }

    #[test]
    fn mean_matches_batch_formula() {
        let mut rng = rand_chacha::ChaCha8Rng::seed_from_u64(42);
        let mut m = Mean::new();
        let mut data = Vec::new();
        for _ in 0..1000 {
            let x = rand::Rng::gen_range(&mut rng, -100.0..100.0);
            m.update(x).unwrap();
            data.push(x);
        }
        let batch: f64 = data.iter().sum::<f64>() / data.len() as f64;
        assert!((m.value() - batch).abs() < 1e-9);
    }
}