arael 0.5.2

Nonlinear optimization framework with compile-time symbolic differentiation
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

use arael::model::{Param, Model};
use arael::vect::vect3f;
use arael::matrix::matrix3f;
use arael::info;

// A simple data point — no optimizable parameters
#[arael::model]
struct DataEntry {
    x: f32,
    y: f32,
}

// Linear model y = a*x + b with two optimizable parameters
#[arael::model]
#[arael(skip_self_block)]
struct LinearModel {
    a: Param<f32>,
    b: Param<f32>,
    data: Vec<DataEntry>,
}

fn main() {
    // ===== LinearModel demo =====
    info!("=== LinearModel: serialize / update / deserialize ===");
    info!("");

    let mut model = LinearModel {
        a: Param::new(1.0),
        b: Param::new(0.0),
        data: vec![
            DataEntry { x: 1.0, y: 2.1 },
            DataEntry { x: 2.0, y: 3.9 },
            DataEntry { x: 3.0, y: 6.1 },
        ],
    };

    // Step 1: serialize — collects all optimizable parameter values
    let mut params = Vec::new();
    model.serialize_params32(&mut params);
    info!("Serialized params: {:?}", params);  // [1.0, 0.0]
    info!("  a index={}, b index={}", model.a.value, model.b.value);

    // Step 2: update_self — initialize work values from current values
    model.update_self();
    info!("");
    info!("After update_self:");
    info!("  a.work={}, b.work={}", model.a.work(), model.b.work());

    // Step 3: simulate optimizer modifying the parameter vector
    params[0] = 2.0;   // a = 2.0
    params[1] = 0.1;   // b = 0.1
    info!("");
    info!("Optimizer sets params to: {:?}", params);

    // Step 4: update — copies new values from data into work
    model.update32(&params);
    info!("After update:");
    info!("  a.work={}, b.work={}", model.a.work(), model.b.work());
    info!("  a.value={}, b.value={} (unchanged)", model.a.value, model.b.value);

    // Evaluate cost with work values
    let cost: f32 = model.data.iter().map(|d| {
        let predicted = model.a.work() * d.x + model.b.work();
        let residual = d.y - predicted;
        residual * residual
    }).sum();
    info!("  Cost = {:.4}", cost);

    // Step 5: deserialize — commit optimized values back
    model.deserialize_params32(&params);
    info!("");
    info!("After deserialize (commit):");
    info!("  a.value={}, b.value={}", model.a.value, model.b.value);

    // ===== Fixed parameters demo =====
    info!("");
    info!("=== Fixed parameter demo ===");
    info!("");

    let mut model2 = LinearModel {
        a: Param::fixed(2.0),  // fixed: will not be optimized
        b: Param::new(0.0),    // optimizable
        data: vec![DataEntry { x: 1.0, y: 2.0 }],
    };

    let mut params2 = Vec::new();
    model2.serialize_params32(&mut params2);
    info!("With a=fixed(2.0), b=new(0.0):");
    info!("  Serialized params: {:?} (only b)", params2);

    params2[0] = 0.5;
    model2.update32(&params2);
    info!("  After update: a.work={}, b.work={}", model2.a.work(), model2.b.work());

    // ===== Computed field demo =====
    info!("");
    info!("=== Computed field (rotation from euler angles) ===");
    info!("");

    #[arael::model]
    #[arael(skip_self_block)]
    struct Pose {
        pos: Param<vect3f>,
        ea: Param<vect3f>,
        #[arael(compute = ea.rotation_matrix())]
        mr2w: matrix3f,
    }

    let mut pose = Pose {
        pos: Param::new(vect3f::new(1.0, 2.0, 3.0)),
        ea: Param::new(vect3f::new(0.0, 0.0, std::f32::consts::FRAC_PI_2)),
        mr2w: matrix3f::identity(),
    };

    let mut params3 = Vec::new();
    pose.serialize_params32(&mut params3);
    info!("Pose params: {:?}", params3);  // [1, 2, 3, 0, 0, pi/2]
    info!("  {} params total (pos=3 + ea=3)", params3.len());

    // update_self: computes mr2w from ea.work()
    pose.update_self();
    info!("");
    info!("After update_self, mr2w computed from euler angles:");
    info!("  mr2w = {:?}", pose.mr2w);

    // Simulate optimizer changing euler angles
    params3[5] = 0.0;  // set ea.z = 0 (no rotation)
    pose.update32(&params3);
    info!("");
    info!("After update with ea.z=0:");
    info!("  ea.work = {:?}", pose.ea.work());
    info!("  mr2w = {:?} (should be ~identity)", pose.mr2w);
}