diffsol 0.17.0

A library for solving ordinary differential equations (ODEs) in Rust.
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

use std::marker::PhantomData;

use crate::{Matrix, Op, Vector};
use std::cell::RefCell;

use super::BuilderOp;

/// An [`Op`] wrapping a user-provided initial-condition closure with `std::autodiff` support.
///
/// When the `autodiff` feature is enabled, [`call_func`](Self::call_func) is annotated
/// with `#[autodiff_reverse]` to automatically generate `call_sens_vjp`.
pub struct ConstantClosureAutodiff<M: Matrix, F> {
    func: F,
    nout: usize,
    nparams: usize,
    tmp_nstates: RefCell<M::V>,
    tmp_nstates2: RefCell<M::V>,
    ctx: M::C,
    _phantom: PhantomData<M>,
}

impl<M: Matrix, F> ConstantClosureAutodiff<M, F> {
    pub fn new(func: F, nout: usize, nparams: usize, ctx: M::C) -> Self {
        Self {
            func,
            nout,
            nparams,
            tmp_nstates: RefCell::new(M::V::zeros(0, ctx.clone())),
            tmp_nstates2: RefCell::new(M::V::zeros(0, ctx.clone())),
            ctx,
            _phantom: PhantomData,
        }
    }
}

impl<M: Matrix, F> BuilderOp for ConstantClosureAutodiff<M, F> {
    fn calculate_sparsity(&mut self, _y0: &Self::V, _t0: Self::T, _p: &Self::V) {}
    fn set_nstates(&mut self, nstates: usize) {
        self.tmp_nstates = RefCell::new(M::V::zeros(nstates, self.ctx.clone()));
        self.tmp_nstates2 = RefCell::new(M::V::zeros(nstates, self.ctx.clone()));
    }
    fn set_nout(&mut self, nout: usize) {
        self.nout = nout;
    }
    fn set_nparams(&mut self, nparams: usize) {
        self.nparams = nparams;
    }
}

impl<M: Matrix, F> Op for ConstantClosureAutodiff<M, F> {
    type V = M::V;
    type T = M::T;
    type M = M;
    type C = M::C;
    fn nstates(&self) -> usize {
        0
    }
    fn nout(&self) -> usize {
        self.nout
    }
    fn nparams(&self) -> usize {
        self.nparams
    }
    fn context(&self) -> &Self::C {
        &self.ctx
    }
}

#[cfg(feature = "autodiff")]
mod autodiff_impl {
    use super::*;
    use crate::{ConstantOp, ConstantOpSensAdjoint, ParameterisedOp, Vector};
    use std::autodiff::autodiff_reverse;

    impl<M: Matrix, F: Fn(&[M::T], M::T, &mut [M::T])> ConstantClosureAutodiff<M, F> {
        #[autodiff_reverse(call_sens_vjp, Const, Duplicated, Const, Duplicated)]
        pub fn call_func(&self, p: &[M::T], t: M::T, y: &mut [M::T]) {
            (self.func)(p, t, y)
        }
    }

    impl<M: Matrix, F: Fn(&[M::T], M::T, &mut [M::T])> ConstantOp
        for ParameterisedOp<'_, ConstantClosureAutodiff<M, F>>
    {
        fn call_inplace(&self, t: Self::T, y: &mut Self::V) {
            y.for_each_batch_host([self.p], |y, [p], _| self.op.call_func(p, t, y));
        }
    }

    impl<M: Matrix, F: Fn(&[M::T], M::T, &mut [M::T])> ConstantOpSensAdjoint
        for ParameterisedOp<'_, ConstantClosureAutodiff<M, F>>
    {
        fn sens_transpose_mul_inplace(&self, t: Self::T, v: &Self::V, y: &mut Self::V) {
            let mut tmp_nstates = self.op.tmp_nstates.borrow_mut();
            let mut tmp_nstates2 = self.op.tmp_nstates2.borrow_mut();
            <M::V as Vector>::for_each_batch_mut_host(
                [y, &mut tmp_nstates, &mut tmp_nstates2],
                [v, self.p],
                |[y, tmp_nstates, tmp_nstates2], [v, p], _| {
                    tmp_nstates.copy_from_slice(v);
                    self.op.call_sens_vjp(p, y, t, tmp_nstates2, tmp_nstates);
                    for y in y.iter_mut() {
                        *y = -*y;
                    }
                },
            );
        }
    }
}

#[cfg(test)]
mod tests {
    use crate::{
        context::nalgebra::NalgebraContext, BuilderOp, ConstantOp, ConstantOpSensAdjoint,
        NalgebraMat, NalgebraVec, ParameterisedOp, Vector,
    };

    use super::ConstantClosureAutodiff;

    type M = NalgebraMat<f64>;
    type V = NalgebraVec<f64>;

    fn initial_condition(p: &[f64], t: f64, y: &mut [f64]) {
        y[0] = p[0] * p[0] + 2.0 * p[1] + t;
        y[1] = 3.0 * p[0] + p[1] - t;
    }

    #[test]
    fn autodiff_constant_closure_generates_parameter_gradient() {
        let ctx = NalgebraContext::default();
        let mut op = ConstantClosureAutodiff::<M, _>::new(initial_condition, 2, 2, ctx);
        op.set_nstates(2);
        let p = V::from_vec(vec![2.0, 5.0], ctx);
        let pop = ParameterisedOp::new(&op, &p);

        let mut value = V::zeros(2, ctx);
        pop.call_inplace(0.5, &mut value);
        value.assert_eq_st(&V::from_vec(vec![14.5, 10.5], ctx), 1e-12);

        let seed = V::from_vec(vec![7.0, 11.0], ctx);
        let mut parameter_vjp = V::zeros(2, ctx);
        pop.sens_transpose_mul_inplace(0.5, &seed, &mut parameter_vjp);
        parameter_vjp.assert_eq_st(&V::from_vec(vec![-61.0, -25.0], ctx), 1e-12);
    }
}