basin 2.0.0

Numerical optimization in pure Rust, with pluggable linear-algebra backends and WASM support.
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

//! Counted evaluation and model construction shared by the NLLS drivers.

use super::{LossFunction, RobustLeastSquares};
use crate::core::constraint::RelaxedProblem;
use crate::core::constraint::evaluation::BoxProblem;
use crate::core::math::{Scalar, ScaleRowsInPlace, VectorIndex, VectorLen};
use crate::core::problem::{Jacobian, Problem, Residual};
use crate::core::state::PointState;

pub(crate) type NllsStep<V, F, E> =
    Result<crate::SolverStep<PointState<V, F>>, E>;

// Separate implementations keep additional robust math capabilities off the
// existing solver bounds, including downstream minimal Jacobian types.
pub(crate) trait Evaluation<V, M, F: Scalar> {
    type Error;
    const ROBUST: bool;
    fn residual(&mut self, x: &V) -> Result<V, Self::Error>;
    fn jacobian(&mut self, x: &V) -> Result<M, Self::Error>;
    fn residual_and_jacobian(&mut self, x: &V) -> Result<(V, M), Self::Error>;
    fn cost(&self, r: &V, quadratic: impl FnOnce(&V) -> F) -> F;
    // The inner None retains raw residuals without cloning on quadratic paths.
    fn model(&self, r: &V, j: M) -> Option<(Option<V>, M)>;
    fn valid_vector(&self, vector: &V) -> bool;
}

pub(crate) trait BoundedEvaluation<V, M, F: Scalar>:
    Evaluation<V, M, F>
{
    fn lower(&self) -> &V;
    fn upper(&self) -> &V;
}

impl<P, V: Clone, M, F: Scalar> Evaluation<V, M, F> for Problem<P>
where
    P: Residual<Param = V, Output = V> + Jacobian<Jacobian = M>,
{
    type Error = P::Error;
    const ROBUST: bool = false;
    fn residual(&mut self, x: &V) -> Result<V, Self::Error> {
        Problem::residual(self, x)
    }
    fn jacobian(&mut self, x: &V) -> Result<M, Self::Error> {
        Problem::jacobian(self, x)
    }
    fn residual_and_jacobian(&mut self, x: &V) -> Result<(V, M), Self::Error> {
        Problem::residual_and_jacobian(self, x)
    }
    fn cost(&self, r: &V, quadratic: impl FnOnce(&V) -> F) -> F {
        quadratic(r)
    }
    fn model(&self, _r: &V, j: M) -> Option<(Option<V>, M)> {
        Some((None, j))
    }
    fn valid_vector(&self, _: &V) -> bool {
        true
    }
}

// Relaxing a box must retain the loss, scale, and raw callback accounting.
macro_rules! robust_evaluation {
    ($problem:ty, $($field:ident).*) => {
        impl<P, L, V, M, F> Evaluation<V, M, F> for Problem<$problem>
        where
            F: Scalar,
            P: Residual<Param = V, Output = V> + Jacobian<Jacobian = M>,
            L: LossFunction<F>,
            V: Clone + VectorLen + VectorIndex<F>,
            M: ScaleRowsInPlace<F>,
        {
            type Error = P::Error;
            const ROBUST: bool = true;
            fn residual(&mut self, x: &V) -> Result<V, Self::Error> {
                self.counts_mut().residual_evals += 1;
                self.inner()$(.$field)*.problem.residual(x)
            }
            fn jacobian(&mut self, x: &V) -> Result<M, Self::Error> {
                self.counts_mut().jacobian_evals += 1;
                self.inner()$(.$field)*.problem.jacobian(x)
            }
            fn residual_and_jacobian(
                &mut self,
                x: &V,
            ) -> Result<(V, M), Self::Error> {
                self.counts_mut().residual_evals += 1;
                self.counts_mut().jacobian_evals += 1;
                self.inner()$(.$field)*.problem.residual_and_jacobian(x)
            }
            fn cost(&self, r: &V, _: impl FnOnce(&V) -> F) -> F {
                self.inner()$(.$field)*.residual_cost(r)
            }
            fn model(&self, r: &V, mut j: M) -> Option<(Option<V>, M)> {
                let (residual, factors) =
                    self.inner()$(.$field)*.model_residual(r)?;
                j.scale_rows_in_place(&factors);
                Some((Some(residual), j))
            }
            fn valid_vector(&self, v: &V) -> bool {
                (0..v.vec_len()).all(|i| v.get_scalar(i).is_finite())
            }
        }
    };
}
robust_evaluation!(RobustLeastSquares<P, L, F>,);
robust_evaluation!(RelaxedProblem<'_, RobustLeastSquares<P, L, F>>, problem);

impl<P, V: Clone, M, F: Scalar> BoundedEvaluation<V, M, F> for Problem<P>
where
    P: Residual<Param = V, Output = V>
        + Jacobian<Jacobian = M>
        + BoxProblem<Param = V>,
{
    fn lower(&self) -> &V {
        self.inner().lower()
    }
    fn upper(&self) -> &V {
        self.inner().upper()
    }
}

impl<P, L, V, M, F> BoundedEvaluation<V, M, F>
    for Problem<RobustLeastSquares<P, L, F>>
where
    F: Scalar,
    P: Residual<Param = V, Output = V>
        + Jacobian<Jacobian = M>
        + BoxProblem<Param = V>,
    L: LossFunction<F>,
    V: Clone + VectorLen + VectorIndex<F>,
    M: ScaleRowsInPlace<F>,
{
    fn lower(&self) -> &V {
        self.inner().problem.lower()
    }
    fn upper(&self) -> &V {
        self.inner().problem.upper()
    }
}