aprender-core 0.29.2

Next-generation machine learning library in pure 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
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

pub(crate) use super::*;

#[test]
fn test_softmax_temperature_basic() {
    let logits = vec![1.0, 2.0, 3.0];

    // T=1 (standard softmax)
    let soft1 = softmax_temperature(&logits, 1.0);
    assert!((soft1.iter().sum::<f64>() - 1.0).abs() < 1e-10);

    // T=2 (softer)
    let soft2 = softmax_temperature(&logits, 2.0);
    assert!((soft2.iter().sum::<f64>() - 1.0).abs() < 1e-10);

    // Higher temperature = more uniform
    let variance1: f64 = soft1.iter().map(|x| (x - 1.0 / 3.0).powi(2)).sum();
    let variance2: f64 = soft2.iter().map(|x| (x - 1.0 / 3.0).powi(2)).sum();
    assert!(variance2 < variance1);
}

#[test]
fn test_softmax_temperature_extreme() {
    let logits = vec![0.0, 10.0];

    // Very low temperature (approaches argmax)
    let hard = softmax_temperature(&logits, 0.1);
    assert!(hard[1] > 0.99);

    // Very high temperature (approaches uniform)
    let uniform = softmax_temperature(&logits, 100.0);
    assert!((uniform[0] - uniform[1]).abs() < 0.1);
}

#[test]
fn test_softmax_empty() {
    let result = softmax_temperature(&[], 1.0);
    assert!(result.is_empty());
}

#[test]
fn test_kl_divergence_same() {
    let p = vec![0.25, 0.25, 0.25, 0.25];
    let kl = kl_divergence(&p, &p);
    assert!(kl.abs() < 1e-10);
}

#[test]
fn test_kl_divergence_different() {
    let p = vec![0.9, 0.1];
    let q = vec![0.5, 0.5];
    let kl = kl_divergence(&p, &q);
    assert!(kl > 0.0);
}

#[test]
fn test_cross_entropy() {
    // Perfect prediction
    let probs = vec![0.0, 1.0, 0.0];
    let labels = vec![0.0, 1.0, 0.0];
    let ce = cross_entropy(&probs, &labels);
    assert!(ce < 0.01);

    // Bad prediction
    let probs_bad = vec![0.9, 0.05, 0.05];
    let ce_bad = cross_entropy(&probs_bad, &labels);
    assert!(ce_bad > ce);
}

#[test]
fn test_binary_cross_entropy() {
    // Perfect prediction
    let bce = binary_cross_entropy(1.0, 1.0);
    assert!(bce < 0.01);

    // Wrong prediction
    let bce_bad = binary_cross_entropy(0.1, 1.0);
    assert!(bce_bad > 1.0);
}

#[test]
fn test_soft_target_generator() {
    let generator = SoftTargetGenerator::with_temperature(3.0);
    let logits = vec![1.0, 2.0, 3.0];
    let soft = generator.generate(&logits);

    assert_eq!(soft.len(), 3);
    assert!((soft.iter().sum::<f64>() - 1.0).abs() < 1e-10);
}

#[test]
fn test_soft_target_generator_batch() {
    let generator = SoftTargetGenerator::with_temperature(2.0);
    let logits = vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0]; // 2 samples × 3 classes
    let soft = generator.generate_batch(&logits, 3);

    assert_eq!(soft.len(), 6);

    // Each sample should sum to 1
    assert!((soft[0..3].iter().sum::<f64>() - 1.0).abs() < 1e-10);
    assert!((soft[3..6].iter().sum::<f64>() - 1.0).abs() < 1e-10);
}

#[test]
fn test_distillation_loss_compute() {
    let loss = DistillationLoss::new();

    let student_logits = vec![1.0, 2.0, 3.0];
    let teacher_logits = vec![1.0, 2.5, 2.8];
    let hard_labels = vec![0.0, 0.0, 1.0];

    let result = loss.compute(&student_logits, &teacher_logits, &hard_labels);
    assert!(result.is_ok());
    assert!(result.unwrap().is_finite());
}

#[test]
fn test_distillation_loss_gradient() {
    let loss = DistillationLoss::new();

    let student_logits = vec![1.0, 2.0, 3.0];
    let teacher_logits = vec![1.0, 2.5, 2.8];
    let hard_labels = vec![0.0, 0.0, 1.0];

    let grad = loss.gradient(&student_logits, &teacher_logits, &hard_labels);
    assert!(grad.is_ok());
    assert_eq!(grad.unwrap().len(), 3);
}

#[test]
fn test_distillation_loss_dimension_mismatch() {
    let loss = DistillationLoss::new();

    let student_logits = vec![1.0, 2.0];
    let teacher_logits = vec![1.0, 2.0, 3.0];
    let hard_labels = vec![0.0, 1.0];

    assert!(loss
        .compute(&student_logits, &teacher_logits, &hard_labels)
        .is_err());
}

#[test]
fn test_linear_distiller_forward() {
    let distiller = LinearDistiller::new(3, 2);

    let features = vec![1.0, 2.0, 3.0];
    let logits = distiller.forward(&features);

    assert!(logits.is_ok());
    assert_eq!(logits.unwrap().len(), 2);
}

#[test]
fn test_linear_distiller_train_step() {
    let config = DistillationConfig {
        learning_rate: 0.1,
        ..Default::default()
    };
    let mut distiller = LinearDistiller::with_config(2, 2, config);

    let features = vec![1.0, 0.5];
    let teacher_logits = vec![0.5, 1.5];
    let hard_labels = vec![0.0, 1.0];

    let loss1 = distiller
        .train_step(&features, &teacher_logits, &hard_labels)
        .unwrap();

    // Train more
    for _ in 0..10 {
        distiller
            .train_step(&features, &teacher_logits, &hard_labels)
            .unwrap();
    }

    let loss2 = distiller
        .train_step(&features, &teacher_logits, &hard_labels)
        .unwrap();

    // Loss should decrease
    assert!(loss2 < loss1 * 1.5); // Allow some variance
}

#[test]
fn test_linear_distiller_predict() {
    let mut distiller = LinearDistiller::new(2, 3);

    // Train briefly
    let features = vec![1.0, 0.0];
    let teacher_logits = vec![0.0, 0.0, 5.0]; // Strongly predict class 2
    let hard_labels = vec![0.0, 0.0, 1.0];

    for _ in 0..50 {
        distiller
            .train_step(&features, &teacher_logits, &hard_labels)
            .unwrap();
    }

    let pred = distiller.predict(&features).unwrap();
    assert_eq!(pred, 2);
}

#[test]
fn test_linear_distiller_dimension_mismatch() {
    let distiller = LinearDistiller::new(3, 2);

    let features = vec![1.0, 2.0]; // Wrong size
    assert!(distiller.forward(&features).is_err());
}

#[test]
fn test_distillation_config_default() {
    let config = DistillationConfig::default();
    assert_eq!(config.temperature, DEFAULT_TEMPERATURE);
    assert_eq!(config.alpha, DEFAULT_ALPHA);
}

#[test]
fn test_distillation_config_builder() {
    let config = DistillationConfig::default()
        .with_temperature(5.0)
        .with_alpha(0.9);

    assert_eq!(config.temperature, 5.0);
    assert_eq!(config.alpha, 0.9);
}

#[test]
fn test_soft_target_generator_default() {
    let generator = SoftTargetGenerator::default();
    assert_eq!(generator.temperature, DEFAULT_TEMPERATURE);
}

#[test]
fn test_distillation_loss_default() {
    let loss = DistillationLoss::default();
    assert_eq!(loss.config().temperature, DEFAULT_TEMPERATURE);
}

#[test]
fn test_temperature_effect_on_dark_knowledge() {
    let logits = vec![1.0, 3.0, 2.0];

    // Low temperature - peaked distribution
    let low_t = softmax_temperature(&logits, 1.0);
    // High temperature - flatter distribution
    let high_t = softmax_temperature(&logits, 5.0);

    // The smallest class probability should be higher with higher T
    let min_low = low_t.iter().cloned().fold(f64::INFINITY, f64::min);
    let min_high = high_t.iter().cloned().fold(f64::INFINITY, f64::min);

    assert!(
        min_high > min_low,
        "Higher T should reveal more dark knowledge"
    );
}