nuts-rs 0.16.1

Sample from unnormalized densities using Hamiltonian MCMC
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

use crate::{
    hamiltonian::{DivergenceInfo, Point},
    math_base::Math,
    nuts::{Collector, NutsOptions},
    state::State,
};

/// Settings for step size adaptation
#[derive(Debug, Clone, Copy)]
pub struct DualAverageOptions {
    pub k: f64,
    pub t0: f64,
    pub gamma: f64,
}

impl Default for DualAverageOptions {
    fn default() -> DualAverageOptions {
        DualAverageOptions {
            k: 0.75,
            t0: 10.,
            gamma: 0.05,
        }
    }
}

#[derive(Clone)]
pub struct DualAverage {
    log_step: f64,
    log_step_adapted: f64,
    hbar: f64,
    mu: f64,
    count: u64,
    settings: DualAverageOptions,
}

impl DualAverage {
    pub fn new(settings: DualAverageOptions, initial_step: f64) -> DualAverage {
        DualAverage {
            log_step: initial_step.ln(),
            log_step_adapted: initial_step.ln(),
            hbar: 0.,
            mu: (10. * initial_step).ln(),
            count: 1,
            settings,
        }
    }

    pub fn advance(&mut self, accept_stat: f64, target: f64) {
        let w = 1. / (self.count as f64 + self.settings.t0);
        self.hbar = (1. - w) * self.hbar + w * (target - accept_stat);
        self.log_step = self.mu - self.hbar * (self.count as f64).sqrt() / self.settings.gamma;
        let mk = (self.count as f64).powf(-self.settings.k);
        self.log_step_adapted = mk * self.log_step + (1. - mk) * self.log_step_adapted;
        self.count += 1;
    }

    pub fn current_step_size(&self) -> f64 {
        self.log_step.exp()
    }

    pub fn current_step_size_adapted(&self) -> f64 {
        self.log_step_adapted.exp()
    }

    #[allow(dead_code)]
    pub fn reset(&mut self, initial_step: f64, bias_factor: f64) {
        self.log_step = initial_step.ln();
        self.log_step_adapted = initial_step.ln();
        self.hbar = 0f64;
        self.mu = (bias_factor * initial_step).ln();
        self.count = 1;
    }
}

pub(crate) struct RunningMean {
    sum: f64,
    count: u64,
}

impl RunningMean {
    fn new() -> RunningMean {
        RunningMean { sum: 0., count: 0 }
    }

    fn add(&mut self, value: f64) {
        self.sum += value;
        self.count += 1;
    }

    pub(crate) fn current(&self) -> f64 {
        self.sum / self.count as f64
    }

    pub(crate) fn reset(&mut self) {
        self.sum = 0f64;
        self.count = 0;
    }

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

pub struct AcceptanceRateCollector {
    initial_energy: f64,
    pub(crate) mean: RunningMean,
    pub(crate) mean_sym: RunningMean,
}

impl AcceptanceRateCollector {
    pub(crate) fn new() -> AcceptanceRateCollector {
        AcceptanceRateCollector {
            initial_energy: 0.,
            mean: RunningMean::new(),
            mean_sym: RunningMean::new(),
        }
    }
}

impl<M: Math, P: Point<M>> Collector<M, P> for AcceptanceRateCollector {
    fn register_leapfrog(
        &mut self,
        _math: &mut M,
        _start: &State<M, P>,
        end: &State<M, P>,
        divergence_info: Option<&DivergenceInfo>,
    ) {
        match divergence_info {
            Some(_) => {
                self.mean.add(0.);
                self.mean_sym.add(0.);
            }
            None => {
                let base_energy = self.initial_energy;
                let other_energy = end.energy();

                let diff = base_energy - other_energy;
                self.mean.add(diff.min(0.).exp());
                self.mean_sym
                    .add(2. * diff.min(0.).exp() / (1. + diff.exp()));
            }
        };
    }

    fn register_init(&mut self, _math: &mut M, state: &State<M, P>, _options: &NutsOptions) {
        self.initial_energy = state.energy();
        self.mean.reset();
        self.mean_sym.reset();
    }
}