flodl 0.2.1

floDl — a flow-graph deep learning framework built on libtorch
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

use crate::autograd::{Variable, conv_transpose1d};
use crate::tensor::{Result, Device};

use super::init::{kaiming_uniform, uniform_bias};
use super::parameter::Parameter;
use super::Module;

/// Transposed 1D convolution (deconvolution) layer.
///
/// Weight shape: `[in_channels, out_channels / groups, kernel_size]`.
/// Bias shape: `[out_channels]` (optional).
pub struct ConvTranspose1d {
    pub weight: Parameter,
    pub bias: Option<Parameter>,
    pub stride: i64,
    pub padding: i64,
    pub output_padding: i64,
    pub dilation: i64,
    pub groups: i64,
}

impl ConvTranspose1d {
    /// Create a ConvTranspose1d layer with default settings and bias.
    pub fn new(
        in_channels: i64, out_channels: i64, kernel_size: i64,
    ) -> Result<Self> {
        Self::build(in_channels, out_channels, kernel_size, true, 1, 0, 0, 1, 1, Device::CPU)
    }

    /// Create a ConvTranspose1d layer on a specific device.
    pub fn on_device(
        in_channels: i64, out_channels: i64, kernel_size: i64, device: Device,
    ) -> Result<Self> {
        Self::build(in_channels, out_channels, kernel_size, true, 1, 0, 0, 1, 1, device)
    }

    /// Fully configurable ConvTranspose1d constructor.
    #[allow(clippy::too_many_arguments)]
    pub fn build(
        in_channels: i64, out_channels: i64, kernel_size: i64,
        with_bias: bool,
        stride: i64, padding: i64, output_padding: i64,
        dilation: i64, groups: i64, device: Device,
    ) -> Result<Self> {
        // Note: weight shape is [in_channels, out_channels/groups, K]
        let shape = [in_channels, out_channels / groups, kernel_size];
        let fan_in = in_channels * kernel_size;

        let weight_data = kaiming_uniform(&shape, fan_in, 5.0_f64.sqrt(), device)?;
        let weight = Variable::new(weight_data, true);

        let bias = if with_bias {
            let bias_data = uniform_bias(fan_in, &[out_channels], device)?;
            Some(Parameter {
                variable: Variable::new(bias_data, true),
                name: "bias".into(),
            })
        } else {
            None
        };

        Ok(ConvTranspose1d {
            weight: Parameter { variable: weight, name: "weight".into() },
            bias,
            stride,
            padding,
            output_padding,
            dilation,
            groups,
        })
    }
}

impl Module for ConvTranspose1d {
    fn name(&self) -> &str { "conv_t1d" }

    fn forward(&self, input: &Variable) -> Result<Variable> {
        conv_transpose1d(
            input,
            &self.weight.variable,
            self.bias.as_ref().map(|b| &b.variable),
            self.stride,
            self.padding,
            self.output_padding,
            self.dilation,
            self.groups,
        )
    }

    fn parameters(&self) -> Vec<Parameter> {
        let mut params = vec![self.weight.clone()];
        if let Some(ref b) = self.bias {
            params.push(b.clone());
        }
        params
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::tensor::{Tensor, test_device, test_opts};

    #[test]
    fn test_conv_transpose1d_forward() {
        let conv = ConvTranspose1d::on_device(4, 2, 3, test_device()).unwrap();
        let x = Variable::new(
            Tensor::randn(&[1, 4, 10], test_opts()).unwrap(), false,
        );
        let y = conv.forward(&x).unwrap();
        // output_size = (10-1)*1 - 2*0 + 3 = 12
        assert_eq!(y.shape(), vec![1, 2, 12]);
    }

    #[test]
    fn test_conv_transpose1d_gradient() {
        let conv = ConvTranspose1d::on_device(4, 2, 3, test_device()).unwrap();
        let x = Variable::new(
            Tensor::randn(&[1, 4, 8], test_opts()).unwrap(), true,
        );
        let y = conv.forward(&x).unwrap().sum().unwrap();
        y.backward().unwrap();
        assert!(x.grad().is_some());
        assert!(conv.weight.variable.grad().is_some());
    }

    #[test]
    fn test_conv_transpose1d_parameters() {
        let conv = ConvTranspose1d::new(4, 2, 3).unwrap();
        assert_eq!(conv.parameters().len(), 2); // weight + bias
    }

    #[test]
    fn test_conv_transpose1d_with_stride() {
        let conv = ConvTranspose1d::build(
            4, 2, 3, true, 2, 0, 0, 1, 1, test_device(),
        ).unwrap();
        let x = Variable::new(
            Tensor::randn(&[1, 4, 5], test_opts()).unwrap(), false,
        );
        let y = conv.forward(&x).unwrap();
        // output_size = (5-1)*2 - 2*0 + 3 = 11
        assert_eq!(y.shape(), vec![1, 2, 11]);
    }
}