aprender-core 0.31.2

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
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280

#[cfg(test)]
mod tests {
    use super::super::*;
    use crate::optim::Optimizer;
    use crate::primitives::Vector;

    #[test]
    fn test_sgd_basic_update() {
        let mut optimizer = SGD::new(0.1);
        let mut params = Vector::from_slice(&[1.0, 2.0]);
        let gradients = Vector::from_slice(&[0.5, 1.0]);

        optimizer.step(&mut params, &gradients);

        assert!((params[0] - 0.95).abs() < 1e-6);
        assert!((params[1] - 1.9).abs() < 1e-6);
    }

    #[test]
    fn test_sgd_with_momentum() {
        let mut optimizer = SGD::new(0.1).with_momentum(0.9);
        let mut params = Vector::from_slice(&[0.0]);
        let gradients = Vector::from_slice(&[1.0]);

        // First step: v = 0.9*0 + 0.1*1 = 0.1, params = 0 - 0.1 = -0.1
        optimizer.step(&mut params, &gradients);
        assert!((params[0] - (-0.1)).abs() < 1e-6);

        // Second step: v = 0.9*0.1 + 0.1*1 = 0.19, params = -0.1 - 0.19 = -0.29
        optimizer.step(&mut params, &gradients);
        assert!((params[0] - (-0.29)).abs() < 1e-6);
    }

    #[test]
    fn test_adam_basic_update() {
        let mut optimizer = Adam::new(0.001);
        let mut params = Vector::from_slice(&[1.0, 2.0]);
        let gradients = Vector::from_slice(&[0.1, 0.2]);

        let original = params.clone();
        optimizer.step(&mut params, &gradients);

        // Parameters should have decreased (gradients are positive)
        assert!(params[0] < original[0]);
        assert!(params[1] < original[1]);
        assert_eq!(optimizer.steps(), 1);
    }

    #[test]
    fn test_adam_reset() {
        let mut optimizer = Adam::new(0.001);
        let mut params = Vector::from_slice(&[1.0]);
        let gradients = Vector::from_slice(&[1.0]);

        optimizer.step(&mut params, &gradients);
        assert_eq!(optimizer.steps(), 1);

        optimizer.reset();
        assert_eq!(optimizer.steps(), 0);
    }

    // ================================================================
    // Additional coverage tests for missed branches
    // ================================================================

    #[test]
    fn test_sgd_reset() {
        let mut optimizer = SGD::new(0.1).with_momentum(0.9);
        let mut params = Vector::from_slice(&[1.0, 2.0]);
        let gradients = Vector::from_slice(&[0.5, 0.5]);

        optimizer.step(&mut params, &gradients);
        // Velocity is now initialized
        assert!(optimizer.velocity.is_some());

        optimizer.reset();
        assert!(optimizer.velocity.is_none());
    }

    #[test]
    fn test_sgd_has_momentum() {
        let sgd_no_momentum = SGD::new(0.1);
        assert!(!sgd_no_momentum.has_momentum());

        let sgd_with_momentum = SGD::new(0.1).with_momentum(0.9);
        assert!(sgd_with_momentum.has_momentum());

        let sgd_zero_momentum = SGD::new(0.1).with_momentum(0.0);
        assert!(!sgd_zero_momentum.has_momentum());
    }

    #[test]
    fn test_sgd_learning_rate_accessor() {
        let sgd = SGD::new(0.05);
        assert!((sgd.learning_rate() - 0.05).abs() < 1e-6);
    }

    #[test]
    fn test_sgd_momentum_accessor() {
        let sgd = SGD::new(0.1).with_momentum(0.99);
        assert!((sgd.momentum() - 0.99).abs() < 1e-6);
    }

    #[test]
    fn test_sgd_no_momentum_multi_step() {
        let mut optimizer = SGD::new(0.1);
        let mut params = Vector::from_slice(&[5.0]);
        let gradients = Vector::from_slice(&[1.0]);

        // Step 1: 5.0 - 0.1*1.0 = 4.9
        optimizer.step(&mut params, &gradients);
        assert!((params[0] - 4.9).abs() < 1e-6);

        // Step 2: 4.9 - 0.1*1.0 = 4.8
        optimizer.step(&mut params, &gradients);
        assert!((params[0] - 4.8).abs() < 1e-6);
    }

    #[test]
    fn test_sgd_momentum_velocity_reinit_on_size_change() {
        let mut optimizer = SGD::new(0.1).with_momentum(0.9);

        // Step with 2 params
        let mut params2 = Vector::from_slice(&[1.0, 2.0]);
        let grads2 = Vector::from_slice(&[0.1, 0.2]);
        optimizer.step(&mut params2, &grads2);

        // Step with 3 params (different size triggers velocity reinitialization)
        let mut params3 = Vector::from_slice(&[1.0, 2.0, 3.0]);
        let grads3 = Vector::from_slice(&[0.1, 0.2, 0.3]);
        optimizer.step(&mut params3, &grads3);

        // Should not panic; velocity was reinitialized
        assert!(optimizer.velocity.is_some());
    }

    #[test]
    fn test_adam_custom_hyperparameters() {
        let optimizer = Adam::new(0.0001)
            .with_beta1(0.95)
            .with_beta2(0.9999)
            .with_epsilon(1e-7);

        assert!((optimizer.learning_rate() - 0.0001).abs() < 1e-9);
        assert!((optimizer.beta1() - 0.95).abs() < 1e-9);
        assert!((optimizer.beta2() - 0.9999).abs() < 1e-9);
        assert!((optimizer.epsilon() - 1e-7).abs() < 1e-15);
    }

    #[test]
    fn test_adam_multiple_steps() {
        let mut optimizer = Adam::new(0.01);
        let mut params = Vector::from_slice(&[1.0, 2.0, 3.0]);
        let gradients = Vector::from_slice(&[0.1, 0.2, 0.3]);

        for _ in 0..10 {
            optimizer.step(&mut params, &gradients);
        }

        assert_eq!(optimizer.steps(), 10);
        // Params should have decreased
        assert!(params[0] < 1.0);
        assert!(params[1] < 2.0);
        assert!(params[2] < 3.0);
    }

    #[test]
    fn test_adam_reset_clears_moments() {
        let mut optimizer = Adam::new(0.001);
        let mut params = Vector::from_slice(&[1.0]);
        let gradients = Vector::from_slice(&[1.0]);

        optimizer.step(&mut params, &gradients);
        assert!(optimizer.m.is_some());
        assert!(optimizer.v.is_some());
        assert_eq!(optimizer.t, 1);

        optimizer.reset();
        assert!(optimizer.m.is_none());
        assert!(optimizer.v.is_none());
        assert_eq!(optimizer.t, 0);
    }

    #[test]
    fn test_adam_size_change_reinit() {
        let mut optimizer = Adam::new(0.001);

        // Step with 2 params
        let mut params2 = Vector::from_slice(&[1.0, 2.0]);
        let grads2 = Vector::from_slice(&[0.1, 0.2]);
        optimizer.step(&mut params2, &grads2);
        assert_eq!(optimizer.steps(), 1);

        // Step with 3 params (triggers moment reinitialization)
        let mut params3 = Vector::from_slice(&[1.0, 2.0, 3.0]);
        let grads3 = Vector::from_slice(&[0.1, 0.2, 0.3]);
        optimizer.step(&mut params3, &grads3);

        // Step counter resets to 1 after reinit
        assert_eq!(optimizer.steps(), 1);
    }

    #[test]
    fn test_sgd_optimizer_trait() {
        let mut optimizer = SGD::new(0.1);
        let mut params = Vector::from_slice(&[1.0, 2.0]);
        let gradients = Vector::from_slice(&[0.5, 1.0]);

        // Use Optimizer trait methods
        Optimizer::step(&mut optimizer, &mut params, &gradients);
        assert!((params[0] - 0.95).abs() < 1e-6);

        Optimizer::reset(&mut optimizer);
        // After reset, velocity is None
        assert!(optimizer.velocity.is_none());
    }

    #[test]
    fn test_adam_optimizer_trait() {
        let mut optimizer = Adam::new(0.001);
        let mut params = Vector::from_slice(&[1.0, 2.0]);
        let gradients = Vector::from_slice(&[0.1, 0.2]);

        // Use Optimizer trait methods
        Optimizer::step(&mut optimizer, &mut params, &gradients);
        assert_eq!(optimizer.steps(), 1);

        Optimizer::reset(&mut optimizer);
        assert_eq!(optimizer.steps(), 0);
    }

    #[test]
    fn test_sgd_clone() {
        let optimizer = SGD::new(0.05).with_momentum(0.9);
        let cloned = optimizer.clone();

        assert!((cloned.learning_rate() - 0.05).abs() < 1e-6);
        assert!((cloned.momentum() - 0.9).abs() < 1e-6);
    }

    #[test]
    fn test_adam_clone() {
        let optimizer = Adam::new(0.001).with_beta1(0.95).with_beta2(0.9999);
        let cloned = optimizer.clone();

        assert!((cloned.learning_rate() - 0.001).abs() < 1e-9);
        assert!((cloned.beta1() - 0.95).abs() < 1e-9);
        assert!((cloned.beta2() - 0.9999).abs() < 1e-9);
    }

    #[test]
    fn test_sgd_debug() {
        let optimizer = SGD::new(0.1);
        let debug_str = format!("{:?}", optimizer);
        assert!(debug_str.contains("SGD"));
        assert!(debug_str.contains("learning_rate"));
    }

    #[test]
    fn test_adam_debug() {
        let optimizer = Adam::new(0.001);
        let debug_str = format!("{:?}", optimizer);
        assert!(debug_str.contains("Adam"));
        assert!(debug_str.contains("learning_rate"));
    }

    #[test]
    fn test_adam_zero_gradient() {
        let mut optimizer = Adam::new(0.001);
        let mut params = Vector::from_slice(&[1.0, 2.0]);
        let gradients = Vector::from_slice(&[0.0, 0.0]);

        let original = params.clone();
        optimizer.step(&mut params, &gradients);

        // With zero gradients, params should barely change (only due to bias correction of zeros)
        assert!((params[0] - original[0]).abs() < 1e-4);
        assert!((params[1] - original[1]).abs() < 1e-4);
    }
}