autograd 0.8.0

Tensors and differentiable operations in Rust
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

extern crate ndarray;

use ndarray_ext;
use ndarray_ext::NdArray;
use op;
use tensor::Tensor;

pub struct Zeros;
pub struct Ones;
pub struct Range;
pub struct ConvertToTensor {
    pub arr: NdArray,
}
pub struct Scalar {
    pub val: f32,
}

impl op::Op for Scalar {
    fn name(&self) -> &str {
        "Scalar"
    }

    fn compute(&self, _: ::runtime::OpComputeContext) -> op::ComputeResult {
        vec![Ok(ndarray::arr0(self.val).into_dyn())]
    }

    fn grad(&self, _: &Tensor, _: &[&Tensor], _: &Tensor) -> Vec<Option<Tensor>> {
        vec![None]
    }
}

impl op::Op for Zeros {
    fn name(&self) -> &str {
        "Zeros"
    }

    fn compute(&self, ctx: ::runtime::OpComputeContext) -> op::ComputeResult {
        let xs = ctx.grab_inputs();
        let shape: &NdArray = xs[0];
        let ret = if let Some(a) = shape.as_slice() {
            Ok(ndarray_ext::zeros(
                a.iter().map(|&b| b as usize).collect::<Vec<_>>().as_slice(),
            ))
        } else {
            Ok(ndarray_ext::zeros(
                shape
                    .iter()
                    .map(|&b| b as usize)
                    .collect::<Vec<_>>()
                    .as_slice(),
            ))
        };
        vec![ret]
    }

    fn grad(&self, _: &Tensor, _: &[&Tensor], _: &Tensor) -> Vec<Option<Tensor>> {
        vec![None]
    }
}

impl op::Op for Ones {
    fn name(&self) -> &str {
        "Ones"
    }

    fn compute(&self, ctx: ::runtime::OpComputeContext) -> op::ComputeResult {
        let xs = ctx.grab_inputs();
        let shape: &NdArray = xs[0];
        let ret = if let Some(a) = shape.as_slice() {
            Ok(ndarray_ext::ones(
                a.iter().map(|&b| b as usize).collect::<Vec<_>>().as_slice(),
            ))
        } else {
            Ok(ndarray_ext::ones(
                shape
                    .iter()
                    .map(|&b| b as usize)
                    .collect::<Vec<_>>()
                    .as_slice(),
            ))
        };
        vec![ret]
    }

    fn grad(&self, _: &Tensor, _: &[&Tensor], _: &Tensor) -> Vec<Option<Tensor>> {
        vec![None]
    }
}

impl op::Op for Range {
    fn name(&self) -> &str {
        "Range"
    }

    fn compute(&self, ctx: ::runtime::OpComputeContext) -> op::ComputeResult {
        let xs = ctx.grab_inputs();
        let x0 = xs[0];
        let x1 = xs[1];
        let x2 = xs[2];

        let true_shape = &[];
        if x0.shape() != true_shape || x1.shape() != true_shape || x2.shape() != true_shape {
            panic!("Inputs to `range` should be 0-ranked tensors");
        }

        let start = x0[ndarray::IxDyn(&[])];
        let end = x1[ndarray::IxDyn(&[])];
        let step = x2[ndarray::IxDyn(&[])];

        if start > end {
            panic!("Start and end of `range` is wrong.");
        }

        vec![Ok(ndarray::Array1::range(start, end, step).into_dyn())]
    }

    fn grad(&self, _: &Tensor, _: &[&Tensor], _: &Tensor) -> Vec<Option<Tensor>> {
        vec![None, None, None]
    }
}

impl op::Op for ConvertToTensor {
    fn name(&self) -> &str {
        "ConvertToTensor"
    }

    fn compute(&self, _: ::runtime::OpComputeContext) -> op::ComputeResult {
        vec![Ok(self.arr.clone())]
    }

    fn grad(&self, _: &Tensor, _: &[&Tensor], _: &Tensor) -> Vec<Option<Tensor>> {
        vec![]
    }
}