aprender-core 0.30.0

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


#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_relu_functional() {
        let x = Tensor::from_slice(&[-1.0, 0.0, 1.0]);
        let y = relu(&x);
        assert_eq!(y.data(), &[0.0, 0.0, 1.0]);
    }

    #[test]
    fn test_leaky_relu_functional() {
        let x = Tensor::from_slice(&[-1.0, 0.0, 1.0]);
        let y = leaky_relu(&x, 0.1);
        assert_eq!(y.data(), &[-0.1, 0.0, 1.0]);
    }

    #[test]
    fn test_softmax_functional() {
        let x = Tensor::new(&[1.0, 2.0, 3.0], &[1, 3]);
        let y = softmax(&x, -1);

        // Should sum to 1
        let sum: f32 = y.data().iter().sum();
        assert!((sum - 1.0).abs() < 1e-5);
    }

    #[test]
    fn test_log_softmax_functional() {
        let x = Tensor::new(&[1.0, 2.0, 3.0], &[1, 3]);
        let y = log_softmax(&x, -1);

        // exp(log_softmax) should sum to 1
        let sum: f32 = y.data().iter().map(|&v| v.exp()).sum();
        assert!((sum - 1.0).abs() < 1e-5);
    }

    #[test]
    fn test_linear_functional() {
        // Identity weight, no bias
        let x = Tensor::new(&[1.0, 2.0, 3.0], &[1, 3]);
        let weight = Tensor::new(&[1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0], &[3, 3]);

        let y = linear(&x, &weight, None);
        assert_eq!(y.data(), &[1.0, 2.0, 3.0]);
    }

    #[test]
    fn test_linear_functional_with_bias() {
        let x = Tensor::new(&[1.0, 2.0], &[1, 2]);
        let weight = Tensor::new(&[1.0, 0.0, 0.0, 1.0], &[2, 2]);
        let bias = Tensor::new(&[10.0, 20.0], &[2]);

        let y = linear(&x, &weight, Some(&bias));
        assert_eq!(y.data(), &[11.0, 22.0]);
    }

    #[test]
    fn test_dropout_eval() {
        let x = Tensor::ones(&[100]);
        let y = dropout(&x, 0.5, false); // eval mode
        assert_eq!(y.data(), x.data());
    }

    // =========================================================================
    // Additional coverage tests for functional interface
    // =========================================================================

    #[test]
    fn test_sigmoid_functional() {
        let x = Tensor::from_slice(&[0.0, 1.0, -1.0]);
        let y = sigmoid(&x);
        // sigmoid(0) = 0.5
        assert!((y.data()[0] - 0.5).abs() < 1e-5);
        // sigmoid(x) is always in (0, 1)
        for &val in y.data() {
            assert!(val > 0.0 && val < 1.0);
        }
    }

    #[test]
    fn test_tanh_functional() {
        let x = Tensor::from_slice(&[0.0, 1.0, -1.0]);
        let y = tanh(&x);
        // tanh(0) = 0
        assert!((y.data()[0]).abs() < 1e-5);
        // tanh is in (-1, 1)
        for &val in y.data() {
            assert!(val >= -1.0 && val <= 1.0);
        }
    }

    #[test]
    fn test_gelu_functional() {
        let x = Tensor::from_slice(&[0.0, 1.0, -1.0]);
        let y = gelu(&x);
        // GELU(0) = 0
        assert!((y.data()[0]).abs() < 1e-5);
        // GELU(1) ≈ 0.841
        assert!((y.data()[1] - 0.841).abs() < 0.01);
        // GELU(-1) ≈ -0.158
        assert!((y.data()[2] - (-0.158)).abs() < 0.02);
    }

    #[test]
    fn test_dropout_training_zeros_some() {
        // With training=true and p=0.5, some values should be zeroed
        let x = Tensor::ones(&[1000]);
        let y = dropout(&x, 0.5, true);

        let zeros = y.data().iter().filter(|&&v| v == 0.0).count();
        let scaled = y
            .data()
            .iter()
            .filter(|&&v| (v - 2.0).abs() < 0.001)
            .count();

        // With p=0.5, roughly half should be zero, half should be scaled by 2
        assert!(
            zeros > 300 && zeros < 700,
            "Expected ~500 zeros, got {zeros}"
        );
        assert!(
            scaled > 300 && scaled < 700,
            "Expected ~500 scaled, got {scaled}"
        );
    }

    #[test]
    fn test_dropout_zero_probability() {
        let x = Tensor::ones(&[100]);
        let y = dropout(&x, 0.0, true); // p=0 means no dropout
        assert_eq!(y.data(), x.data());
    }

    #[test]
    fn test_softmax_multi_batch() {
        let x = Tensor::new(&[1.0, 2.0, 3.0, 4.0, 5.0, 6.0], &[2, 3]);
        let y = softmax(&x, -1);

        // Each row should sum to 1
        let sum1: f32 = y.data()[0..3].iter().sum();
        let sum2: f32 = y.data()[3..6].iter().sum();
        assert!((sum1 - 1.0).abs() < 1e-5);
        assert!((sum2 - 1.0).abs() < 1e-5);
    }

    #[test]
    fn test_log_softmax_multi_batch() {
        let x = Tensor::new(&[1.0, 2.0, 3.0, 4.0, 5.0, 6.0], &[2, 3]);
        let y = log_softmax(&x, -1);

        // exp(log_softmax) should sum to 1 for each row
        let sum1: f32 = y.data()[0..3].iter().map(|v| v.exp()).sum();
        let sum2: f32 = y.data()[3..6].iter().map(|v| v.exp()).sum();
        assert!((sum1 - 1.0).abs() < 1e-5);
        assert!((sum2 - 1.0).abs() < 1e-5);
    }

    #[test]
    fn test_linear_batch() {
        // Test linear with batch dimension
        let x = Tensor::new(&[1.0, 2.0, 3.0, 4.0, 5.0, 6.0], &[2, 3]);
        let weight = Tensor::new(&[1.0, 0.0, 0.0, 0.0, 1.0, 0.0], &[2, 3]);
        let bias = Tensor::new(&[10.0, 20.0], &[2]);

        let y = linear(&x, &weight, Some(&bias));
        // Row 1: [1,2,3] @ [[1,0],[0,1],[0,0]]^T + [10,20] = [1,2] + [10,20] = [11,22]
        assert!((y.data()[0] - 11.0).abs() < 1e-5);
        assert!((y.data()[1] - 22.0).abs() < 1e-5);
    }
}