rv 0.19.2

Random variables
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

#[cfg(feature = "serde1")]
use serde::{Deserialize, Serialize};

use crate::traits::SuffStat;
use nalgebra::{DMatrix, DVector};

#[derive(Debug, Clone, PartialEq)]
#[cfg_attr(feature = "serde1", derive(Serialize, Deserialize))]
#[cfg_attr(feature = "serde1", serde(rename_all = "snake_case"))]
pub struct MvGaussianSuffStat {
    n: usize,
    sum_x: DVector<f64>,
    sum_x_sq: DMatrix<f64>,
}

impl MvGaussianSuffStat {
    #[inline]
    #[must_use]
    pub fn new(dims: usize) -> Self {
        MvGaussianSuffStat {
            n: 0,
            sum_x: DVector::zeros(dims),
            sum_x_sq: DMatrix::zeros(dims, dims),
        }
    }

    /// Create a sufficient statistic from components without checking whether
    /// they are valid.
    ///
    /// # Example
    /// ```rust
    /// use rv::data::MvGaussianSuffStat;
    /// use rv::prelude::SuffStat;
    /// use nalgebra::{DVector, dvector, DMatrix};
    ///
    /// let data: Vec<DVector<f64>> = vec![
    ///   dvector![0.1, 0.2],
    ///   dvector![0.3, 0.4],
    /// ];
    ///
    /// let mut stat_b = MvGaussianSuffStat::new(2);
    /// stat_b.observe_many(&data);
    ///
    /// let n = data.len();
    /// let sum_x: DVector<f64> = data.iter().sum();
    /// let sum_x_sq: DMatrix<f64> = data.iter().map(|x| x * x.transpose()).sum();
    ///
    /// let stat_a = MvGaussianSuffStat::from_parts_unchecked(n, sum_x, sum_x_sq);
    ///
    /// assert_eq!(stat_a.n(), stat_b.n());
    /// assert::close(stat_a.sum_x()[0], stat_b.sum_x()[0], 1e-10);
    /// assert::close(stat_a.sum_x()[1], stat_b.sum_x()[1], 1e-10);
    /// assert::close(stat_a.sum_x_sq()[(0, 0)], stat_b.sum_x_sq()[(0, 0)], 1e-10);
    /// assert::close(stat_a.sum_x_sq()[(0, 1)], stat_b.sum_x_sq()[(0, 1)], 1e-10);
    /// assert::close(stat_a.sum_x_sq()[(1, 0)], stat_b.sum_x_sq()[(1, 0)], 1e-10);
    /// assert::close(stat_a.sum_x_sq()[(1, 1)], stat_b.sum_x_sq()[(1, 1)], 1e-10);
    ///
    /// ```
    #[inline]
    #[must_use]
    pub fn from_parts_unchecked(
        n: usize,
        sum_x: DVector<f64>,
        sum_x_sq: DMatrix<f64>,
    ) -> Self {
        MvGaussianSuffStat { n, sum_x, sum_x_sq }
    }

    /// Get the number of observations
    #[inline]
    #[must_use]
    pub fn n(&self) -> usize {
        self.n
    }

    /// Get the sum of observations
    #[inline]
    #[must_use]
    pub fn sum_x(&self) -> &DVector<f64> {
        &self.sum_x
    }

    /// Get the sum of X^2
    #[inline]
    #[must_use]
    pub fn sum_x_sq(&self) -> &DMatrix<f64> {
        &self.sum_x_sq
    }
}

impl SuffStat<DVector<f64>> for MvGaussianSuffStat {
    fn n(&self) -> usize {
        self.n
    }

    fn observe(&mut self, x: &DVector<f64>) {
        self.n += 1;
        if self.n == 1 {
            self.sum_x = x.clone();
            self.sum_x_sq = x * x.transpose();
        } else {
            self.sum_x += x;
            self.sum_x_sq += x * x.transpose();
        }
    }

    fn forget(&mut self, x: &DVector<f64>) {
        self.n -= 1;
        if self.n > 0 {
            self.sum_x -= x;
            self.sum_x_sq -= x * x.transpose();
        } else {
            let dims = self.sum_x.len();
            self.sum_x = DVector::zeros(dims);
            self.sum_x_sq = DMatrix::zeros(dims, dims);
        }
    }

    fn merge(&mut self, other: Self) {
        self.n += other.n;
        self.sum_x += other.sum_x;
        self.sum_x_sq += other.sum_x_sq;
    }
}

#[cfg(test)]
mod tests {
    use nalgebra::dvector;

    use super::*;

    #[test]
    fn observe_forget() {
        let mut stat = MvGaussianSuffStat::new(2);

        stat.observe(&dvector![0.1, 0.2]);
        stat.observe(&dvector![0.3, 0.4]);

        dbg!(&stat.sum_x_sq);

        assert_eq!(stat.n(), 2);
        assert::close(stat.sum_x()[0], 0.4, 1e-10);
        assert::close(stat.sum_x()[1], 0.6, 1e-10);
        assert::close(stat.sum_x_sq()[(0, 0)], 0.1, 1e-10);
        assert::close(stat.sum_x_sq()[(0, 1)], 0.14, 1e-10);
        assert::close(stat.sum_x_sq()[(1, 0)], 0.14, 1e-10);
        assert::close(stat.sum_x_sq()[(1, 1)], 0.2, 1e-10);

        stat.forget(&dvector![0.1, 0.2]);
        assert_eq!(stat.n(), 1);
        assert::close(stat.sum_x()[0], 0.3, 1e-10);
        assert::close(stat.sum_x()[1], 0.4, 1e-10);
        assert::close(stat.sum_x_sq()[(0, 0)], 0.09, 1e-10);
        assert::close(stat.sum_x_sq()[(0, 1)], 0.12, 1e-10);
        assert::close(stat.sum_x_sq()[(1, 0)], 0.12, 1e-10);
        assert::close(stat.sum_x_sq()[(1, 1)], 0.16, 1e-10);

        stat.forget(&dvector![0.3, 0.4]);
        assert_eq!(stat.n(), 0);
        assert::close(stat.sum_x()[0], 0.0, 1e-10);
        assert::close(stat.sum_x()[1], 0.0, 1e-10);
        assert::close(stat.sum_x_sq()[(0, 0)], 0.0, 1e-10);
        assert::close(stat.sum_x_sq()[(0, 1)], 0.0, 1e-10);
        assert::close(stat.sum_x_sq()[(1, 0)], 0.0, 1e-10);
        assert::close(stat.sum_x_sq()[(1, 1)], 0.0, 1e-10);
    }

    #[test]
    fn merge() {
        let mut a = MvGaussianSuffStat::new(2);
        let mut b = MvGaussianSuffStat::new(2);
        let mut c = MvGaussianSuffStat::new(2);

        a.observe(&dvector![0.1, 0.2]);
        b.observe(&dvector![0.3, 0.4]);

        c.observe(&dvector![0.1, 0.2]);
        c.observe(&dvector![0.3, 0.4]);

        <MvGaussianSuffStat as SuffStat<DVector<f64>>>::merge(&mut a, b);

        assert_eq!(a.n(), c.n());
        assert::close(a.sum_x()[0], c.sum_x()[0], 1e-10);
        assert::close(a.sum_x()[1], c.sum_x()[1], 1e-10);
        assert::close(a.sum_x_sq()[(0, 0)], c.sum_x_sq()[(0, 0)], 1e-10);
        assert::close(a.sum_x_sq()[(0, 1)], c.sum_x_sq()[(0, 1)], 1e-10);
        assert::close(a.sum_x_sq()[(1, 0)], c.sum_x_sq()[(1, 0)], 1e-10);
        assert::close(a.sum_x_sq()[(1, 1)], c.sum_x_sq()[(1, 1)], 1e-10);
    }
}