neuronika 0.2.0

Tensors and dynamic neural networks.
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

use super::{
    assert_almost_equals, new_backward_input, new_input, new_tensor, Backward, Data, Forward,
    Gradient, KLDivLoss, KLDivLossBackward, Reduction,
};
use ndarray::arr0;

#[test]
fn mean() {
    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Forward Pass ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    let target = new_input((2, 3), vec![0.2, 0.5, 0.3, 0.6, 0.0, 0.4]);
    let v: Vec<f32> = vec![0.4, 0.5, 0.1, 0.6, 0.1, 0.3]
        .iter()
        .map(|&el: &f32| el.ln())
        .collect();
    let input = new_input((2, 3), v);
    input.forward();

    let loss = KLDivLoss::new(input, target.clone(), Reduction::Mean);

    loss.forward();
    assert_almost_equals(&*loss.data(), &arr0(0.1530));

    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Backward Pass ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

    let input_diff = new_backward_input((2, 3), vec![0.; 6]);
    let loss_backward = KLDivLossBackward::new(input_diff.clone(), target, Reduction::Mean);

    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Seed Gradient ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    *loss_backward.gradient_mut() = arr0(1.);

    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    loss_backward.backward();
    assert_almost_equals(
        &*input_diff.gradient(),
        &new_tensor(
            (2, 3),
            vec![-0.1000, -0.2500, -0.1500, -0.3000, 0.0000, -0.2000],
        ),
    );

    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 2nd Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    loss_backward.backward();
    assert_almost_equals(
        &*input_diff.gradient(),
        &(&new_tensor(
            (2, 3),
            vec![-0.1000, -0.2500, -0.1500, -0.3000, 0.0000, -0.2000],
        ) * 2.),
    );
}

#[test]
fn sum() {
    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Forward Pass ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    let target = new_input((2, 3), vec![0.2, 0.5, 0.3, 0.6, 0.0, 0.4]);
    let v: Vec<f32> = vec![0.4, 0.5, 0.1, 0.6, 0.1, 0.3]
        .iter()
        .map(|&el: &f32| el.ln())
        .collect();
    let input = new_input((2, 3), v);
    input.forward();

    let loss = KLDivLoss::new(input, target.clone(), Reduction::Sum);

    loss.forward();
    assert_almost_equals(&*loss.data(), &arr0(0.3060));

    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Backward Pass ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

    let input_diff = new_backward_input((2, 3), vec![0.; 6]);
    let loss_backward = KLDivLossBackward::new(input_diff.clone(), target, Reduction::Sum);

    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Seed Gradient ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    *loss_backward.gradient_mut() = arr0(1.);

    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    loss_backward.backward();
    assert_almost_equals(
        &*input_diff.gradient(),
        &new_tensor(
            (2, 3),
            vec![-0.2000, -0.5000, -0.3000, -0.6000, 0.0000, -0.4000],
        ),
    );

    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 2nd Evaluation ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    loss_backward.backward();
    assert_almost_equals(
        &*input_diff.gradient(),
        &(&new_tensor(
            (2, 3),
            vec![-0.2000, -0.5000, -0.3000, -0.6000, 0.0000, -0.4000],
        ) * 2.),
    );
}

#[test]
fn debug_forward() {
    let target = new_input((2, 3), vec![0.2, 0.5, 0.3, 0.6, 0.0, 0.4]);
    let v: Vec<f32> = vec![0.4, 0.5, 0.1, 0.6, 0.1, 0.3]
        .iter()
        .map(|&el: &f32| el.ln())
        .collect();
    let input = new_input((2, 3), v);

    let loss = KLDivLoss::new(input, target.clone(), Reduction::Mean);

    let output = "KLDivLoss { data: 0.0, shape=[], strides=[], layout=CFcf (0xf), const ndim=0, reduction: Mean, computed: false }";

    assert_eq!(output, format!("{:?}", loss));
}

#[test]
fn display_forward() {
    let target = new_input((2, 3), vec![0.2, 0.5, 0.3, 0.6, 0.0, 0.4]);
    let v: Vec<f32> = vec![0.4, 0.5, 0.1, 0.6, 0.1, 0.3]
        .iter()
        .map(|&el: &f32| el.ln())
        .collect();
    let input = new_input((2, 3), v);

    let loss = KLDivLoss::new(input, target.clone(), Reduction::Mean);

    assert_eq!(format!("{}", loss.data()), format!("{}", loss));
}

#[test]
fn debug_backward() {
    let target = new_input((2, 3), vec![0.2, 0.5, 0.3, 0.6, 0.0, 0.4]);
    let input_diff = new_backward_input((2, 3), vec![0.; 6]);
    let loss = KLDivLossBackward::new(input_diff.clone(), target, Reduction::Mean);

    let output = "KLDivLossBackward { gradient: Some(0.0, shape=[], strides=[], layout=CFcf (0xf), const ndim=0), reduction: Mean, overwrite: true }";

    assert_eq!(output, format!("{:?}", loss));
}

#[test]
fn display_backward() {
    let target = new_input((2, 3), vec![0.2, 0.5, 0.3, 0.6, 0.0, 0.4]);
    let input_diff = new_backward_input((2, 3), vec![0.; 6]);
    let loss = KLDivLossBackward::new(input_diff.clone(), target, Reduction::Mean);

    assert_eq!(format!("{}", loss.gradient()), format!("{}", loss));
}

#[test]
fn no_grad() {
    // KLDivLossBackward
    let node = KLDivLossBackward::new(
        new_backward_input(3, vec![0.; 3]),
        new_input(3, vec![0.; 3]),
        Reduction::Mean,
    );

    node.no_grad();
    assert!(node.gradient.borrow().is_none());

    node.with_grad();
    assert_eq!(&*node.gradient(), arr0(0.));
}