primal-barracuda 0.4.0

barraCuda — BARrier-free Rust Abstracted Cross-platform Unified Dimensional Algebra. Vendor-agnostic GPU compute via WGSL.
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
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
// SPDX-License-Identifier: AGPL-3.0-or-later
#![expect(clippy::unwrap_used, reason = "tests")]

use super::*;
use crate::device::WgpuDevice;
use crate::device::test_pool::{
    get_test_device_if_f64_gpu_available_sync, get_test_device_if_gpu_available_sync,
};
use crate::surrogate::kernels::RBFKernel;
use std::sync::Arc;

fn device() -> Option<Arc<WgpuDevice>> {
    get_test_device_if_f64_gpu_available_sync()
}

fn any_device() -> Option<Arc<WgpuDevice>> {
    get_test_device_if_gpu_available_sync()
}

/// Construct a minimal `RBFSurrogate` from parts for testing predict
/// validation paths without needing GPU compute for training.
fn fake_surrogate(device: Arc<WgpuDevice>, n_dim: usize) -> RBFSurrogate {
    let n_train = 3;
    RBFSurrogate::from_parts(
        device,
        RbfTrainingData {
            train_x: vec![0.0; n_train * n_dim],
            train_y: vec![0.0; n_train],
            n_train,
            n_dim,
        },
        RbfTrainedModel {
            weights: vec![1.0; n_train],
            poly_coeffs: vec![0.0; n_dim + 1],
            kernel: RBFKernel::ThinPlateSpline,
            smoothing: 1e-12,
        },
    )
}

#[test]
fn test_rbf_linear_1d() {
    let Some(dev) = device() else { return };
    let x_train = vec![vec![0.0], vec![1.0], vec![2.0]];
    let y_train = vec![0.0, 2.0, 4.0];
    let surrogate =
        RBFSurrogate::train(dev, &x_train, &y_train, RBFKernel::ThinPlateSpline, 1e-12).unwrap();

    // Should interpolate training points exactly
    let y_pred = surrogate.predict(&x_train).unwrap();
    for i in 0..3 {
        assert!(
            (y_pred[i] - y_train[i]).abs() < 1e-10,
            "Failed to interpolate training point {}: pred = {}, true = {}",
            i,
            y_pred[i],
            y_train[i]
        );
    }

    // Test interpolation
    let y_mid = surrogate.predict(&[vec![1.5]]).unwrap();
    assert!(
        (y_mid[0] - 3.0).abs() < 0.1,
        "Poor interpolation at x=1.5: {}",
        y_mid[0]
    );
}

#[test]
fn test_rbf_quadratic_1d() {
    // Approximate y = x²
    let x_train: Vec<Vec<f64>> = (0..5).map(|i| vec![i as f64]).collect();
    let y_train: Vec<f64> = (0..5).map(|i| (i * i) as f64).collect();

    let Some(dev) = device() else { return };
    let surrogate =
        RBFSurrogate::train(dev, &x_train, &y_train, RBFKernel::ThinPlateSpline, 1e-12).unwrap();

    // Should interpolate training points exactly
    let y_pred = surrogate.predict(&x_train).unwrap();
    for i in 0..5 {
        assert!(
            (y_pred[i] - y_train[i]).abs() < 1e-8,
            "Failed at training point x={}: pred = {}, true = {}",
            i,
            y_pred[i],
            y_train[i]
        );
    }
}

#[test]
fn test_rbf_2d() {
    // Simple 2D function: f(x,y) = x + y
    let x_train = vec![
        vec![0.0, 0.0],
        vec![1.0, 0.0],
        vec![0.0, 1.0],
        vec![1.0, 1.0],
    ];
    let y_train = vec![0.0, 1.0, 1.0, 2.0];

    let Some(dev) = device() else { return };
    let surrogate =
        RBFSurrogate::train(dev, &x_train, &y_train, RBFKernel::ThinPlateSpline, 1e-12).unwrap();

    // Test center point
    let y_center = surrogate.predict(&[vec![0.5, 0.5]]).unwrap();
    assert!(
        (y_center[0] - 1.0).abs() < 0.1,
        "Poor interpolation at center: {}",
        y_center[0]
    );
}

#[test]
fn test_rbf_gaussian_kernel() {
    let x_train = vec![vec![0.0], vec![1.0], vec![2.0]];
    let y_train = vec![0.0, 1.0, 0.0]; // Peak at x=1

    let Some(dev) = device() else { return };
    let surrogate = RBFSurrogate::train(
        dev,
        &x_train,
        &y_train,
        RBFKernel::Gaussian { epsilon: 1.0 },
        1e-12,
    )
    .unwrap();

    // Should interpolate training points
    let y_pred = surrogate.predict(&x_train).unwrap();
    for i in 0..3 {
        assert!((y_pred[i] - y_train[i]).abs() < 1e-8);
    }
}

#[test]
fn test_rbf_empty_training_data() {
    let Some(dev) = any_device() else { return };
    let result = RBFSurrogate::train(dev, &[], &[], RBFKernel::ThinPlateSpline, 1e-12);
    assert!(result.is_err());
}

#[test]
fn test_rbf_mismatched_lengths() {
    let Some(dev) = any_device() else { return };
    let x_train = vec![vec![0.0], vec![1.0]];
    let y_train = vec![0.0, 1.0, 2.0];
    let result = RBFSurrogate::train(dev, &x_train, &y_train, RBFKernel::ThinPlateSpline, 1e-12);
    assert!(result.is_err());
}

#[test]
fn test_loo_cv_rmse() {
    // With smoothing, LOO-CV should give meaningful results
    let x_train = vec![vec![0.0], vec![0.5], vec![1.0], vec![1.5], vec![2.0]];
    // Noisy linear function: y ≈ 2x
    let y_train = vec![0.1, 1.1, 1.9, 3.1, 3.9];

    // Use moderate smoothing so LOO-CV is defined
    let Some(dev) = device() else { return };
    let surrogate =
        RBFSurrogate::train(dev, &x_train, &y_train, RBFKernel::ThinPlateSpline, 1e-6).unwrap();

    let loo_rmse = surrogate.loo_cv_rmse().unwrap();

    // Should be non-negative
    assert!(loo_rmse >= 0.0);

    // Should be small since data is nearly linear
    assert!(loo_rmse < 1.0, "LOO-CV RMSE too large: {loo_rmse}");
}

#[test]
fn test_loo_cv_errors() {
    let x_train = vec![vec![0.0], vec![1.0], vec![2.0]];
    let y_train = vec![0.0, 1.0, 4.0];

    let Some(dev) = device() else { return };
    let surrogate =
        RBFSurrogate::train(dev, &x_train, &y_train, RBFKernel::ThinPlateSpline, 1e-4).unwrap();

    let errors = surrogate.loo_cv_errors().unwrap();

    // Should have one error per training point
    assert_eq!(errors.len(), 3);

    // Errors should be finite
    for e in &errors {
        assert!(e.is_finite(), "Non-finite LOO error: {e}");
    }
}

#[test]
fn test_loo_cv_with_exact_interpolation() {
    // With very small smoothing (exact interpolation), LOO residuals
    // may be near zero or undefined (H_ii ≈ 1)
    let x_train = vec![vec![0.0], vec![1.0], vec![2.0]];
    let y_train = vec![0.0, 1.0, 4.0];

    let Some(dev) = device() else { return };
    let surrogate =
        RBFSurrogate::train(dev, &x_train, &y_train, RBFKernel::ThinPlateSpline, 1e-12).unwrap();

    // Should not panic
    let _ = surrogate.loo_cv_rmse();
}

#[test]
fn test_rbf_accessors() {
    // Need at least n_dim + 1 points for polynomial augmentation
    let x_train = vec![
        vec![0.0, 0.0],
        vec![1.0, 0.0],
        vec![0.0, 1.0],
        vec![1.0, 1.0],
    ];
    let y_train = vec![0.0, 1.0, 1.0, 2.0];

    let Some(dev) = device() else { return };
    let surrogate =
        RBFSurrogate::train(dev, &x_train, &y_train, RBFKernel::ThinPlateSpline, 1e-12).unwrap();

    assert_eq!(surrogate.n_train(), 4);
    assert_eq!(surrogate.n_dim(), 2);
}

#[test]
fn test_loo_cv_hat_diagonal_correct() {
    // Test that H_ii < 1 when smoothing > 0 (the bug was H_ii = 1 always)
    // Use non-linear noisy data that won't be perfectly fit
    let x_train = vec![
        vec![0.0],
        vec![0.5],
        vec![1.0],
        vec![1.5],
        vec![2.0],
        vec![2.5],
        vec![3.0],
    ];
    // Noisy quadratic: y ≈ x² + noise
    let y_train = vec![0.1, 0.35, 0.9, 2.3, 4.1, 6.2, 9.1];

    let Some(dev) = device() else { return };
    let surrogate = RBFSurrogate::train(
        dev,
        &x_train,
        &y_train,
        RBFKernel::Gaussian { epsilon: 1.0 },
        0.5,
    )
    .unwrap();

    let errors = surrogate.loo_cv_errors().unwrap();

    // With the CORRECT hat matrix and high smoothing, predictions != targets
    // So residuals (y - ŷ) should be non-zero
    // And with H_ii < 1, the LOO residuals should also be non-zero

    // First verify we have some residuals (smoothed predictions differ from targets)
    let train_points: Vec<Vec<f64>> = x_train;
    let predictions = surrogate.predict(&train_points).unwrap();
    let max_residual: f64 = predictions
        .iter()
        .zip(y_train.iter())
        .map(|(p, y)| (p - y).abs())
        .fold(0.0, f64::max);

    // With high smoothing, the fit should be imperfect
    assert!(
        max_residual > 0.01,
        "Expected imperfect fit with high smoothing. Max residual: {max_residual}"
    );

    // LOO errors should be non-zero (the main test)
    let max_abs_error = errors.iter().map(|e| e.abs()).fold(0.0, f64::max);
    assert!(
        max_abs_error > 0.001,
        "LOO errors should be non-zero with smoothing. Max error: {max_abs_error}"
    );
}

#[test]
fn test_loo_cv_smoothing_effect() {
    // More smoothing should generally increase LOO-CV RMSE (underfitting)
    // Less smoothing should decrease LOO-CV RMSE (better fit, but risk overfitting)
    let x_train: Vec<Vec<f64>> = (0..10).map(|i| vec![i as f64 * 0.1]).collect();
    let y_train: Vec<f64> = x_train
        .iter()
        .map(|x| 2.0f64.mul_add(x[0], 0.1 * (x[0] * 10.0).sin()))
        .collect();

    let Some(dev) = device() else { return };
    let kernel = RBFKernel::Gaussian { epsilon: 1.0 };
    let surrogate_low = RBFSurrogate::train(dev.clone(), &x_train, &y_train, kernel, 1e-4).unwrap();
    let surrogate_high = RBFSurrogate::train(dev, &x_train, &y_train, kernel, 0.5).unwrap();

    let rmse_low = surrogate_low.loo_cv_rmse().unwrap();
    let rmse_high = surrogate_high.loo_cv_rmse().unwrap();

    // Both should be finite and positive
    assert!(rmse_low.is_finite() && rmse_low > 0.0);
    assert!(rmse_high.is_finite() && rmse_high > 0.0);

    // High smoothing should generally give higher RMSE (underfitting)
    // But this isn't strictly monotonic, so we just verify they're different
    assert!(
        (rmse_low - rmse_high).abs() > 1e-6,
        "LOO-CV should be sensitive to smoothing. low={rmse_low}, high={rmse_high}"
    );
}

// === Tests that work on any GPU (including llvmpipe) ===
// These exercise validation paths that return before GPU compute.

#[test]
fn test_predict_empty_eval_returns_empty() {
    let Some(dev) = any_device() else { return };
    let surrogate = fake_surrogate(dev, 2);
    let result = surrogate.predict(&[]).unwrap();
    assert!(result.is_empty());
}

#[test]
fn test_predict_dimension_mismatch() {
    let Some(dev) = any_device() else { return };
    let surrogate = fake_surrogate(dev, 2);
    let result = surrogate.predict(&[vec![1.0, 2.0, 3.0]]);
    assert!(result.is_err());
}

#[test]
fn test_accessors_from_parts() {
    let Some(dev) = any_device() else { return };
    let surrogate = fake_surrogate(dev, 3);
    assert_eq!(surrogate.n_train(), 3);
    assert_eq!(surrogate.n_dim(), 3);
}

#[test]
fn test_loo_cv_empty_training_data() {
    let Some(dev) = any_device() else { return };
    let surrogate = RBFSurrogate::from_parts(
        dev,
        RbfTrainingData {
            train_x: vec![],
            train_y: vec![],
            n_train: 0,
            n_dim: 1,
        },
        RbfTrainedModel {
            weights: vec![],
            poly_coeffs: vec![0.0, 0.0],
            kernel: RBFKernel::ThinPlateSpline,
            smoothing: 1e-6,
        },
    );
    let errors = surrogate.loo_cv_errors().unwrap();
    assert!(errors.is_empty());
}

#[test]
fn test_loo_optimal_smoothing_empty_grid() {
    let Some(dev) = any_device() else { return };
    let x = vec![vec![0.0], vec![1.0], vec![2.0]];
    let y = vec![0.0, 1.0, 4.0];
    let result = loo_cv_optimal_smoothing(dev, &x, &y, RBFKernel::ThinPlateSpline, Some(&[]));
    assert!(result.is_err());
}

#[test]
fn test_loo_optimal_smoothing() {
    let Some(dev) = device() else { return };
    let x: Vec<Vec<f64>> = (0..8).map(|i| vec![i as f64 * 0.5]).collect();
    let y: Vec<f64> = x.iter().map(|v| 2.0f64.mul_add(v[0], 0.1)).collect();
    let result = loo_cv_optimal_smoothing(dev, &x, &y, RBFKernel::ThinPlateSpline, None).unwrap();
    assert!(result.smoothing > 0.0);
    assert!(result.rmse >= 0.0);
    assert!(!result.grid_results.is_empty());
}

#[test]
fn test_loo_optimal_smoothing_custom_grid() {
    let Some(dev) = device() else { return };
    let x = vec![vec![0.0], vec![1.0], vec![2.0], vec![3.0]];
    let y = vec![0.0, 1.0, 4.0, 9.0];
    let grid = [1e-8, 1e-4, 1e-2];
    let result =
        loo_cv_optimal_smoothing(dev, &x, &y, RBFKernel::ThinPlateSpline, Some(&grid)).unwrap();
    assert!(result.grid_results.len() <= 3);
    assert!(result.rmse.is_finite());
}

#[test]
fn test_rbf_training_data_struct() {
    let data = RbfTrainingData {
        train_x: vec![1.0, 2.0, 3.0, 4.0],
        train_y: vec![10.0, 20.0],
        n_train: 2,
        n_dim: 2,
    };
    assert_eq!(data.n_train, 2);
    assert_eq!(data.n_dim, 2);
    assert_eq!(data.train_x.len(), 4);
}

#[test]
fn test_rbf_trained_model_struct() {
    let model = RbfTrainedModel {
        weights: vec![1.0, 2.0],
        poly_coeffs: vec![0.5, 0.3, 0.1],
        kernel: RBFKernel::Cubic,
        smoothing: 1e-6,
    };
    assert_eq!(model.weights.len(), 2);
    assert_eq!(model.poly_coeffs.len(), 3);
    assert_eq!(model.kernel.name(), "Cubic");
}

#[test]
fn test_loo_smoothing_struct() {
    let s = LooSmoothing {
        smoothing: 1e-4,
        rmse: 0.01,
        grid_results: vec![(1e-6, 0.1), (1e-4, 0.01), (1e-2, 0.05)],
    };
    assert_eq!(s.smoothing, 1e-4);
    assert_eq!(s.rmse, 0.01);
    assert_eq!(s.grid_results.len(), 3);
}