odesign 0.1.9

odesign is an optimal design of experiments library written in pure 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

mod a_opt;
mod c_opt;
mod costs_opt;
mod d_opt;
use crate::{MatrixDRows, NLPFunctionTarget};
pub use a_opt::AOptimality;
pub use c_opt::COptimality;
pub use costs_opt::CostsOptimality;
pub use d_opt::DOptimality;
use faer::Mat;
use rayon::prelude::*;
use std::sync::Arc;

/// Vector of optimalities
pub type Optimalities<const D: usize> = Vec<Arc<dyn Optimality<D> + Send + Sync>>;

/// Defines a list of optimality measures and their weights.
#[derive(Default)]
pub struct OptimalityMeasures {
    measures: Vec<Arc<dyn NLPFunctionTarget + Sync + Send>>,
    weights: Vec<f64>,
}

impl OptimalityMeasures {
    /// Creates new struct with empty list of measures and weights.
    pub fn new() -> Self {
        Self::default()
    }
    /// Adds an optimality measure and its weight.
    pub fn push(&mut self, measure: Arc<dyn NLPFunctionTarget + Sync + Send>, weight: f64) {
        self.measures.push(measure);
        self.weights.push(weight);
    }
}

impl NLPFunctionTarget for OptimalityMeasures {
    fn val(&self, x: &Mat<f64>) -> f64 {
        match self.measures.len() {
            0 => 0.,
            1 => self.measures[0].val(x),
            2..3 => self
                .measures
                .iter()
                .zip(&self.weights)
                .map(|(m, w)| w * m.val(x))
                .sum(),
            3.. => self
                .measures
                .par_iter()
                .zip(&self.weights)
                .map(|(m, w)| w * m.val(x))
                .sum(),
        }
    }

    fn val_grad(&self, x: &Mat<f64>) -> (f64, Mat<f64>) {
        match self.measures.len() {
            0 => (0.0, Mat::zeros(x.nrows(), 1)),
            1 => self.measures[0].val_grad(x),
            2..3 => self
                .measures
                .iter()
                .zip(&self.weights)
                .map(|(m, w)| {
                    let val_grad = m.val_grad(x);
                    (w * val_grad.0, *w * val_grad.1)
                })
                .reduce(|(mut sum_val, mut sum_grad), (val, grad)| {
                    sum_val += val;
                    sum_grad += grad;
                    (sum_val, sum_grad)
                })
                .unwrap(),
            3.. => self
                .measures
                .par_iter()
                .zip(&self.weights)
                .map(|(m, w)| {
                    let val_grad = m.val_grad(x);
                    (w * val_grad.0, *w * val_grad.1)
                })
                .reduce(
                    || (0.0, Mat::zeros(x.nrows(), 1)),
                    |(mut sum_val, mut sum_grad), (val, grad)| {
                        sum_val += val;
                        sum_grad += grad;
                        (sum_val, sum_grad)
                    },
                ),
        }
    }
    fn val_grad_hes(&self, x: &Mat<f64>) -> (f64, Mat<f64>, Mat<f64>) {
        match self.measures.len() {
            0 => (
                0.0,
                Mat::zeros(x.nrows(), 1),
                Mat::zeros(x.nrows(), x.nrows()),
            ),
            1 => self.measures[0].val_grad_hes(x),
            2..3 => self
                .measures
                .iter()
                .zip(&self.weights)
                .map(|(m, w)| {
                    let val_grad_hes = m.val_grad_hes(x);
                    (w * val_grad_hes.0, *w * val_grad_hes.1, *w * val_grad_hes.2)
                })
                .reduce(
                    |(mut sum_val, mut sum_grad, mut sum_hes), (val, grad, hes)| {
                        sum_val += val;
                        sum_grad += grad;
                        sum_hes += hes;
                        (sum_val, sum_grad, sum_hes)
                    },
                )
                .unwrap(),
            3.. => self
                .measures
                .par_iter()
                .zip(&self.weights)
                .map(|(m, w)| {
                    let val_grad_hes = m.val_grad_hes(x);
                    (w * val_grad_hes.0, *w * val_grad_hes.1, *w * val_grad_hes.2)
                })
                .reduce(
                    || {
                        (
                            0.0,
                            Mat::zeros(x.nrows(), 1),
                            Mat::zeros(x.nrows(), x.nrows()),
                        )
                    },
                    |(mut sum_val, mut sum_grad, mut sum_hes), (val, grad, hes)| {
                        sum_val += val;
                        sum_grad += grad;
                        sum_hes += hes;
                        (sum_val, sum_grad, sum_hes)
                    },
                ),
        }
    }
}

/// Defines an optimality by providing the matrix mean, despersion function and a measure function.
pub trait Optimality<const D: usize> {
    /// A twice continuously differentiable matrix mean function of the optimality.
    fn matrix_mean(&self, supp: Arc<MatrixDRows<D>>) -> Arc<dyn NLPFunctionTarget + Send + Sync>;
    /// A twice continuously differentiable dispersion function of the optimality.
    fn dispersion_function(
        &self,
        supp: Arc<MatrixDRows<D>>,
        weights: Mat<f64>,
        x_id: usize,
    ) -> Arc<dyn NLPFunctionTarget + Send + Sync>;
    /// Returns the optimality measure value.
    fn measure(&self, weights: &Mat<f64>, supp: Arc<MatrixDRows<D>>) -> f64;
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::{Result, assert_nlp_target_consistency};

    struct DesignTest {}

    impl NLPFunctionTarget for DesignTest {
        fn val(&self, _x: &Mat<f64>) -> f64 {
            1.
        }
        fn val_grad(&self, x: &Mat<f64>) -> (f64, Mat<f64>) {
            (self.val(x), Mat::ones(x.nrows(), 1))
        }

        fn val_grad_hes(&self, x: &Mat<f64>) -> (f64, Mat<f64>, Mat<f64>) {
            let val_grad = self.val_grad(x);
            (val_grad.0, val_grad.1, Mat::ones(x.nrows(), x.nrows()))
        }
    }

    #[test]
    fn design_measures() -> Result<()> {
        let dt = Arc::new(DesignTest {});
        for no_measures in 0..42 {
            let mut dms = OptimalityMeasures::new();
            (0..no_measures).for_each(|w| dms.push(dt.clone(), w as f64 + 1.));
            let x = Mat::ones(1, 1);
            let vgh = dms.val_grad_hes(&x);
            let no_measures_f = no_measures as f64;
            let factor = no_measures_f * (no_measures_f + 1.) / 2.;
            let vgh_val = factor;
            let vgh_grad = factor * Mat::<f64>::ones(1, 1);
            let vgh_hes = factor * Mat::<f64>::ones(1, 1);
            assert_eq!(vgh.0, vgh_val);
            assert_eq!(vgh.1, vgh_grad);
            assert_eq!(vgh.2, vgh_hes);
            assert_nlp_target_consistency!(dms, &x);
        }
        Ok(())
    }
}