csrk 1.1.4

Sparse Gaussian Process regression with compactly supported radial kernels
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
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247

#[test]
#[ignore] // remove ignore when you actually want to benchmark
fn timing_kernel_build_naive_vs_hashed() {
    use std::time::Instant;
    use csrk::{GP,PCG64Stream};
    use ndarray::{arr1, Array2};

    // ---- build a 2D grid of training points ----
    let nx = 150;
    let ny = 150;
    let h = 0.2; // spacing < 1.0 so neighbors fall inside support

    let mut x_train_vec: Vec<Vec<f64>> = Vec::new();
    for i in 0..nx {
        for j in 0..ny {
            x_train_vec.push(vec![i as f64 * h, j as f64 * h]);
        }
    }
    // Flatten for array construction
    let x_train_flat: Vec<f64> = x_train_vec.iter()
        .flat_map(|row| row.iter().copied())
        .collect();

    let nsample = x_train_vec.len();
    let ndim = 2;

    println!("points: {}", nsample);
    let x_train = Array2::from_shape_vec((nsample, ndim), x_train_flat).unwrap();

    // dummy data (kernel build only depends on geometry + errors)
    let y_train = arr1(&vec![0.0; nsample]);
    let y_err = arr1(&vec![0.0; nsample]);
    let whitenoise = 1e-10;
    // kernel parameters (pick one representative case)
    let msq_order = 2;

    // =======================================================================
    // ================= Realization Benchmark ===============================
    // =======================================================================
    // build and train full GP
    let scale = arr1(&[1.0, 1.0]);
    let t2 = Instant::now();
    let gp = GP::train(
        x_train.view(),
        y_train.view(),
        y_err.view(),
        scale.view(),
        whitenoise,
        msq_order,
    );
    let t_train = t2.elapsed();
    println!("gp train    : {:?}", t_train);
    // Generate evaluation points
    let mut x_eval_vec = Vec::new();
    for i in 0..nx {
        for j in 0..ny {
            x_eval_vec.push(
                vec![
                    i as f64 * h + 0.05,
                    j as f64 * h + 0.05
                ]
            );
        }
    }
    // Flatten for array construction
    let x_eval_flat: Vec<f64> = x_eval_vec.iter()
        .flat_map(|row| row.iter().copied())
        .collect();
    let neval = x_eval_vec.len();
    let x_eval = Array2::from_shape_vec((neval,ndim),
        x_eval_flat).unwrap();
    println!("eval points : {:?}", neval);
    // Predict mean
    let t3 = Instant::now();
    let mu = gp.predict_mean(x_eval.view());
    let t_mean = t3.elapsed();
    for x in mu.iter() {
        assert!(x.is_finite());
    }
    println!("predict mean: {:?}", t_mean);
    // Realization timing
    let nreal = 200;
    let mut rng = PCG64Stream::new(123456, 987654);
    let t4 = Instant::now();
    let mut acc = 0.0;
    for _ in 0..nreal {
        let r = gp.draw_realization(&mut rng);
        for i in 0..neval {
            acc += r.value(x_eval.row(i))
        }
    }
    let t_real = t4.elapsed();
    println!("eval reals  : {:?}", t_real);
    let total_evals = nreal * neval;
    println!("realizations: {}", nreal);
    println!("total evals: {}", total_evals);
    println!("realization generation time: {:?}", t_real);
    println!("  throughput: {:.3e} evals/sec",
        (total_evals as f64) / t_real.as_secs_f64()
    );
    assert!(acc.is_finite());
}

#[test]
#[ignore]
fn scaling_sweep_kernel_and_realizations() {
    use std::time::Instant;
    use csrk::kernel::{Kernel};
    use csrk::{GP,PCG64Stream};
    use csrk::spatial_hash::SpatialHash;
    use ndarray::{arr1, Array2, ArrayView1};

    let msq_order = 2;
    let whitenoise = 1e-10;
    let h = 0.2;

    println!(
        "(nx,ny),\tntrain,\tnnz,\tnnz/ntrain,\tbuild_s, train_s, mean_s, draw_s, eval_s, evals_per_sec"
    );

    // sweep grid sizes
    let sizes = [30, 50, 75, 100, 125, 150, 200];

    for &n in &sizes {
        let nx = n;
        let ny = n;

        // ---- build training grid ----
        let mut x_train_vec = Vec::new();
        for i in 0..nx {
            for j in 0..ny {
                x_train_vec.push(vec![i as f64 * h, j as f64 * h]);
            }
        }
        // Flatten for array construction
        let x_train_flat: Vec<f64> = x_train_vec.iter()
            .flat_map(|row| row.iter().copied())
            .collect();

        let ndim = 2;
        let nsample = x_train_vec.len();
        let x_train: Array2<f64> = Array2::from_shape_vec((nsample,ndim),
            x_train_flat).unwrap();

        let y_train = arr1(&vec![0.0; nsample]);
        let y_err = arr1(&vec![0.0; nsample]);
        let scale = arr1(&[1.0, 1.0]);

        let mykernel = Kernel::new(
            ndim,
            msq_order,
            scale.view(),
            whitenoise
        );
        let hash = SpatialHash::build(&x_train_vec, 1.0);

        // ---- kernel build ----
        let t0 = Instant::now();
        let k = mykernel.hashed_kernel_construction(
            x_train.view(),
            y_err.view(),
            &hash,
        );
        let build_t = t0.elapsed().as_secs_f64();
        let nnz = k.nnz();

        // ---- train GP----
        let t1 = Instant::now();
        let gp = GP::train(
            x_train.view(),
            y_train.view(),
            y_err.view(),
            scale.view(),
            whitenoise,
            msq_order,
        );

        let train_t = t1.elapsed().as_secs_f64();

        // ---- evaluation points ----
        let mut x_eval_vec = Vec::new();
        for i in 0..nx {
            for j in 0..ny {
                x_eval_vec.push(vec![i as f64 * h + 0.05, j as f64 * h + 0.05]);
            }
        }
        // Flatten for array construction
        let x_eval_flat: Vec<f64> = x_eval_vec.iter()
            .flat_map(|row| row.iter().copied())
            .collect();

        let neval = x_eval_vec.len();
        let x_eval = Array2::from_shape_vec((neval,ndim),
            x_eval_flat).unwrap();
        let nreal_eval = 5;

        // ---- mean timing ----
        let t2 = Instant::now();
        let mean = gp.predict_mean(x_eval.view());
        let mean_t = t2.elapsed().as_secs_f64();
        for mu in mean.iter() {
            assert!(mu.is_finite());
        }

        // ---- realization draw timing ----
        let mut rng = PCG64Stream::new(1234, 5678);

        let nreal_draw = 20;
        let t3 = Instant::now();
        let mut reals = Vec::new();
        for _ in 0..nreal_draw {
            reals.push(gp.draw_realization(&mut rng));
        }
        let draw_t = t3.elapsed().as_secs_f64() / nreal_draw as f64;

        // ---- realization eval timing ----
        let t4 = Instant::now();
        let mut acc = 0.0;
        for r in reals.iter().take(nreal_eval) {
            for x in &x_eval_vec {
                let x_view = ArrayView1::from(x.as_slice());
                acc += r.value(x_view);
            }
        }
        let eval_t = t4.elapsed().as_secs_f64();
        let evals_per_sec = (nreal_eval * neval) as f64 / eval_t;

        // ---- report ----
        println!(
            "({}, {}),\t{},\t{},  \t{:.4},\t{:.4},\t{:.4},\t{:.4},\t{:.4},\t{:.4}\t{:.3e}",
            nx,
            ny,
            nsample,
            nnz,
            nnz as f64 / nsample as f64,
            build_t,
            train_t,
            mean_t,
            draw_t,
            eval_t,
            evals_per_sec,
        );

        assert!(acc.is_finite());
    }
}