native_neural_network_std 0.2.1

Ergonomic std wrapper for the `native_neural_network` crate (no_std) — std-friendly re-exports and utilities.
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

pub struct TensorStd {
    pub data: Vec<f32>,
    pub shape: Vec<usize>,
}

fn elementwise_view_shape(total_len: usize) -> [usize; 5] {
    [1, 1, 1, 1, total_len]
}

impl TensorStd {
    pub fn new(shape: Vec<usize>, data: Vec<f32>) -> Result<Self, &'static str> {
        let mut n = 1usize;
        for &d in &shape {
            n = n.saturating_mul(d);
        }
        if n != data.len() {
            return Err("data length does not match shape");
        }
        Ok(Self { data, shape })
    }

    pub fn zeros(shape: Vec<usize>) -> Self {
        let mut n = 1usize;
        for &d in &shape {
            n = n.saturating_mul(d);
        }
        Self {
            data: vec![0f32; n],
            shape,
        }
    }

    pub fn len(&self) -> usize {
        self.data.len()
    }

    pub fn is_empty(&self) -> bool {
        self.data.is_empty()
    }

    pub fn rank(&self) -> usize {
        self.shape.len()
    }

    pub fn dims(&self) -> &[usize] {
        &self.shape
    }

    pub fn idx_linear(&self, idxs: &[usize]) -> Option<usize> {
        if idxs.len() != self.shape.len() {
            return None;
        }
        let mut linear = 0usize;
        for (i, &ind) in idxs.iter().enumerate() {
            let dim = self.shape[i];
            if ind >= dim {
                return None;
            }
            linear = linear.checked_mul(dim)?;
            linear = linear.checked_add(ind)?;
        }
        if linear < self.data.len() {
            Some(linear)
        } else {
            None
        }
    }

    pub fn get(&self, idxs: &[usize]) -> Option<f32> {
        self.idx_linear(idxs).map(|i| self.data[i])
    }

    pub fn get_mut(&mut self, idxs: &[usize]) -> Option<&mut f32> {
        let i = self.idx_linear(idxs)?;
        Some(&mut self.data[i])
    }

    pub fn as_slice(&self) -> &[f32] {
        &self.data
    }

    pub fn as_mut_slice(&mut self) -> &mut [f32] {
        &mut self.data
    }

    pub fn into_vec(self) -> Vec<f32> {
        self.data
    }
}

pub fn tensor_fill(t: &mut TensorStd, value: f32) {
    t.fill(value);
}

pub fn tensor_scale_in_place(t: &mut TensorStd, scale: f32) {
    t.scale_in_place(scale);
}

pub fn tensor_add_in_place(a: &mut TensorStd, b: &TensorStd) -> bool {
    a.add_in_place(b)
}

impl TensorStd {
    pub fn fill(&mut self, value: f32) {
        let len = self.data.len();
        let mut view = native_neural_network::tensor::TensorView {
            data: self.data.as_mut_slice(),
            shape: elementwise_view_shape(len),
        };
        native_neural_network::tensor::tensor_fill(&mut view, value);
    }

    pub fn tensor_fill(t: &mut TensorStd, value: f32) {
        t.fill(value);
    }

    pub fn scale_in_place(&mut self, scale: f32) {
        let len = self.data.len();
        let mut view = native_neural_network::tensor::TensorView {
            data: self.data.as_mut_slice(),
            shape: elementwise_view_shape(len),
        };
        native_neural_network::tensor::tensor_scale_in_place(&mut view, scale);
    }

    pub fn add_in_place(&mut self, other: &TensorStd) -> bool {
        if self.shape != other.shape || self.data.len() != other.data.len() {
            return false;
        }
        let dst_len = self.data.len();
        let src_len = other.data.len();
        let mut dst = native_neural_network::tensor::TensorView {
            data: self.data.as_mut_slice(),
            shape: elementwise_view_shape(dst_len),
        };
        let mut src_buf = other.data.clone();
        let src = native_neural_network::tensor::TensorView {
            data: src_buf.as_mut_slice(),
            shape: elementwise_view_shape(src_len),
        };
        native_neural_network::tensor::tensor_add_in_place(&mut dst, &src)
    }
}