yscv-autograd 0.1.9

Dynamic computation graph with tape-based reverse-mode autodiff
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

use yscv_kernels::CpuBackend;
use yscv_tensor::Tensor;

use crate::{AutogradError, Graph};

#[test]
fn backward_accumulates_gradient_from_multiple_paths() {
    let mut graph = Graph::new();
    let x = graph.variable(Tensor::from_vec(vec![2], vec![1.0, 2.0]).unwrap());
    let y = graph.add(x, x).unwrap();
    let loss = graph.sum(y).unwrap();
    graph.backward(loss).unwrap();

    let x_grad = graph.grad(x).unwrap().unwrap();
    assert_eq!(x_grad.shape(), &[2]);
    assert_eq!(x_grad.data(), &[2.0, 2.0]);
}

#[test]
fn backward_does_not_produce_grad_for_constant_leaf() {
    let mut graph = Graph::new();
    let x = graph.variable(Tensor::from_vec(vec![2], vec![1.0, 2.0]).unwrap());
    let c = graph.constant(Tensor::from_vec(vec![2], vec![3.0, 4.0]).unwrap());

    let y = graph.add(x, c).unwrap();
    let loss = graph.sum(y).unwrap();
    graph.backward(loss).unwrap();

    assert!(graph.grad(c).unwrap().is_none());
    assert!(graph.grad(x).unwrap().is_some());
}

#[test]
fn backward_requires_scalar_target() {
    let mut graph = Graph::new();
    let x = graph.variable(Tensor::from_vec(vec![2], vec![1.0, 2.0]).unwrap());
    let err = graph.backward(x).unwrap_err();
    assert_eq!(err, AutogradError::NonScalarTarget { shape: vec![2] });
}

#[test]
fn truncate_keeps_only_requested_prefix() {
    let mut graph = Graph::new();
    let _a = graph.variable(Tensor::from_vec(vec![1], vec![1.0]).unwrap());
    let _b = graph.variable(Tensor::from_vec(vec![1], vec![2.0]).unwrap());
    let _c = graph.variable(Tensor::from_vec(vec![1], vec![3.0]).unwrap());
    assert_eq!(graph.node_count(), 3);

    graph.truncate(1).unwrap();
    assert_eq!(graph.node_count(), 1);
}

#[test]
fn truncate_rejects_out_of_range_size() {
    let mut graph = Graph::new();
    let _a = graph.variable(Tensor::from_vec(vec![1], vec![1.0]).unwrap());
    let err = graph.truncate(2).unwrap_err();
    assert_eq!(
        err,
        AutogradError::InvalidTruncate {
            requested: 2,
            available: 1
        }
    );
}

#[test]
fn test_graph_default_no_backend() {
    // Verify existing behavior is unchanged when no backend is set.
    let mut graph = Graph::new();
    let a = graph.variable(Tensor::from_vec(vec![2, 2], vec![1.0, 2.0, 3.0, 4.0]).unwrap());
    let b = graph.variable(Tensor::from_vec(vec![2, 2], vec![5.0, 6.0, 7.0, 8.0]).unwrap());

    let sum = graph.add(a, b).unwrap();
    let sum_val = graph.value(sum).unwrap();
    assert_eq!(sum_val.data(), &[6.0, 8.0, 10.0, 12.0]);

    let mm = graph.matmul_2d(a, b).unwrap();
    let mm_val = graph.value(mm).unwrap();
    assert_eq!(mm_val.shape(), &[2, 2]);
    assert_eq!(mm_val.data(), &[19.0, 22.0, 43.0, 50.0]);

    let r = graph.relu(a).unwrap();
    let r_val = graph.value(r).unwrap();
    assert_eq!(r_val.data(), &[1.0, 2.0, 3.0, 4.0]);
}

#[test]
fn test_graph_with_cpu_backend() {
    // Create a graph with CpuBackend and verify results match default (no backend).
    let mut graph_default = Graph::new();
    let mut graph_backend = Graph::new();
    graph_backend.set_backend(Box::new(CpuBackend));

    let data_a = vec![1.0, 2.0, 3.0, 4.0];
    let data_b = vec![5.0, 6.0, 7.0, 8.0];

    let a1 = graph_default.variable(Tensor::from_vec(vec![2, 2], data_a.clone()).unwrap());
    let b1 = graph_default.variable(Tensor::from_vec(vec![2, 2], data_b.clone()).unwrap());

    let a2 = graph_backend.variable(Tensor::from_vec(vec![2, 2], data_a.clone()).unwrap());
    let b2 = graph_backend.variable(Tensor::from_vec(vec![2, 2], data_b.clone()).unwrap());

    // Test add
    let sum1 = graph_default.add(a1, b1).unwrap();
    let sum2 = graph_backend.add(a2, b2).unwrap();
    assert_eq!(
        graph_default.value(sum1).unwrap().data(),
        graph_backend.value(sum2).unwrap().data(),
    );

    // Test matmul
    let mm1 = graph_default.matmul_2d(a1, b1).unwrap();
    let mm2 = graph_backend.matmul_2d(a2, b2).unwrap();
    assert_eq!(
        graph_default.value(mm1).unwrap().data(),
        graph_backend.value(mm2).unwrap().data(),
    );

    // Test relu
    let r1 = graph_default.relu(a1).unwrap();
    let r2 = graph_backend.relu(a2).unwrap();
    assert_eq!(
        graph_default.value(r1).unwrap().data(),
        graph_backend.value(r2).unwrap().data(),
    );

    // Test sub
    let sub1 = graph_default.sub(a1, b1).unwrap();
    let sub2 = graph_backend.sub(a2, b2).unwrap();
    assert_eq!(
        graph_default.value(sub1).unwrap().data(),
        graph_backend.value(sub2).unwrap().data(),
    );

    // Test mul
    let mul1 = graph_default.mul(a1, b1).unwrap();
    let mul2 = graph_backend.mul(a2, b2).unwrap();
    assert_eq!(
        graph_default.value(mul1).unwrap().data(),
        graph_backend.value(mul2).unwrap().data(),
    );

    // Test sigmoid
    let sig1 = graph_default.sigmoid(a1).unwrap();
    let sig2 = graph_backend.sigmoid(a2).unwrap();
    assert_eq!(
        graph_default.value(sig1).unwrap().data(),
        graph_backend.value(sig2).unwrap().data(),
    );

    // Test softmax
    let sm1 = graph_default.softmax(a1).unwrap();
    let sm2 = graph_backend.softmax(a2).unwrap();
    let sm1_data = graph_default.value(sm1).unwrap().data();
    let sm2_data = graph_backend.value(sm2).unwrap().data();
    for (v1, v2) in sm1_data.iter().zip(sm2_data.iter()) {
        assert!((v1 - v2).abs() < 1e-6, "softmax mismatch: {v1} vs {v2}");
    }

    // Test clear_backend reverts to default behavior
    graph_backend.clear_backend();
    let a3 = graph_backend.variable(Tensor::from_vec(vec![2, 2], data_a).unwrap());
    let b3 = graph_backend.variable(Tensor::from_vec(vec![2, 2], data_b).unwrap());
    let sum3 = graph_backend.add(a3, b3).unwrap();
    assert_eq!(
        graph_default.value(sum1).unwrap().data(),
        graph_backend.value(sum3).unwrap().data(),
    );
}