flodl 0.5.2

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
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
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203

use crate::autograd::{Variable, conv3d};
use crate::tensor::{Result, Device};

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

/// 3D convolution layer.
///
/// Weight shape: `[out_channels, in_channels / groups, kD, kH, kW]`.
/// Bias shape: `[out_channels]` (optional).
///
/// Input: `[batch, in_channels, D, H, W]`.
/// Output: `[batch, out_channels, D_out, H_out, W_out]`.
pub struct Conv3d {
    pub weight: Parameter,
    pub bias: Option<Parameter>,
    pub stride: [i64; 3],
    pub padding: [i64; 3],
    pub dilation: [i64; 3],
    pub groups: i64,
}

/// Builder for configuring Conv3d layers with a fluent API.
pub struct Conv3dBuilder {
    in_channels: i64,
    out_channels: i64,
    kernel_size: [i64; 3],
    with_bias: bool,
    stride: [i64; 3],
    padding: [i64; 3],
    dilation: [i64; 3],
    groups: i64,
    device: Device,
}

impl Conv3dBuilder {
    /// Set the convolution stride (default: [1, 1, 1]). Controls output spatial size per dimension.
    pub fn with_stride(mut self, stride: [i64; 3]) -> Self { self.stride = stride; self }

    /// Set zero-padding added to input (default: [0, 0, 0]). One value per spatial dimension.
    pub fn with_padding(mut self, padding: [i64; 3]) -> Self { self.padding = padding; self }

    /// Set kernel dilation (default: [1, 1, 1]). Increases receptive field without adding parameters.
    pub fn with_dilation(mut self, dilation: [i64; 3]) -> Self { self.dilation = dilation; self }

    /// Set grouped convolution (default: 1). Groups=in_channels gives depthwise convolution.
    pub fn with_groups(mut self, groups: i64) -> Self { self.groups = groups; self }

    /// Disable the bias term.
    pub fn without_bias(mut self) -> Self { self.with_bias = false; self }

    /// Set the target device (default: CPU).
    pub fn on_device(mut self, device: Device) -> Self { self.device = device; self }

    /// Build the convolution layer with the configured parameters.
    pub fn done(self) -> Result<Conv3d> {
        Conv3d::build(
            self.in_channels, self.out_channels, self.kernel_size,
            self.with_bias, self.stride, self.padding, self.dilation,
            self.groups, self.device,
        )
    }
}

impl Conv3d {
    /// Create a Conv3d layer with default stride=1, padding=0, dilation=1, groups=1, with bias.
    pub fn new(
        in_channels: i64, out_channels: i64, kernel_size: [i64; 3],
    ) -> Result<Self> {
        Self::build(in_channels, out_channels, kernel_size, true,
                    [1, 1, 1], [0, 0, 0], [1, 1, 1], 1, Device::CPU)
    }

    /// Create a Conv3d layer on a specific device.
    pub fn on_device(
        in_channels: i64, out_channels: i64, kernel_size: [i64; 3], device: Device,
    ) -> Result<Self> {
        Self::build(in_channels, out_channels, kernel_size, true,
                    [1, 1, 1], [0, 0, 0], [1, 1, 1], 1, device)
    }

    /// Start a fluent builder for full configuration.
    ///
    /// ```ignore
    /// let conv = Conv3d::configure(1, 4, [3, 3, 3])
    ///     .with_padding([1, 1, 1])
    ///     .done()?;
    /// ```
    pub fn configure(in_channels: i64, out_channels: i64, kernel_size: [i64; 3]) -> Conv3dBuilder {
        Conv3dBuilder {
            in_channels, out_channels, kernel_size,
            with_bias: true,
            stride: [1, 1, 1], padding: [0, 0, 0], dilation: [1, 1, 1],
            groups: 1, device: Device::CPU,
        }
    }

    /// Fully configurable Conv3d constructor.
    #[allow(clippy::too_many_arguments)]
    pub fn build(
        in_channels: i64, out_channels: i64, kernel_size: [i64; 3],
        with_bias: bool,
        stride: [i64; 3], padding: [i64; 3], dilation: [i64; 3],
        groups: i64, device: Device,
    ) -> Result<Self> {
        let shape = [out_channels, in_channels / groups,
                     kernel_size[0], kernel_size[1], kernel_size[2]];
        let fan_in = (in_channels / groups) * kernel_size[0] * kernel_size[1] * kernel_size[2];

        let weight_data = kaiming_uniform(&shape, fan_in, 5.0_f64.sqrt(), device)?;

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

        Ok(Conv3d {
            weight: Parameter::new(weight_data, "weight"),
            bias, stride, padding, dilation, groups,
        })
    }
}

impl Module for Conv3d {
    fn name(&self) -> &str { "conv3d" }

    fn forward(&self, input: &Variable) -> Result<Variable> {
        conv3d(
            input, &self.weight.variable,
            self.bias.as_ref().map(|b| &b.variable),
            self.stride, self.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_conv3d_forward() {
        let conv = Conv3d::on_device(1, 4, [3, 3, 3], test_device()).unwrap();
        let x = Variable::new(
            Tensor::randn(&[1, 1, 8, 8, 8], test_opts()).unwrap(),
            false,
        );
        let y = conv.forward(&x).unwrap();
        assert_eq!(y.shape(), vec![1, 4, 6, 6, 6]);
    }

    #[test]
    fn test_conv3d_gradient() {
        let conv = Conv3d::on_device(1, 2, [3, 3, 3], test_device()).unwrap();
        let x = Variable::new(
            Tensor::randn(&[1, 1, 6, 6, 6], test_opts()).unwrap(),
            true,
        );
        let y = conv.forward(&x).unwrap().sum().unwrap();
        y.backward().unwrap();
        assert!(x.grad().is_some());
    }

    #[test]
    fn test_conv3d_no_bias() {
        let conv = Conv3d::configure(1, 4, [3, 3, 3])
            .without_bias()
            .on_device(test_device())
            .done().unwrap();
        assert_eq!(conv.parameters().len(), 1);
        assert!(conv.bias.is_none());
        let x = Variable::new(
            Tensor::randn(&[1, 1, 6, 6, 6], test_opts()).unwrap(), false,
        );
        let y = conv.forward(&x).unwrap();
        assert_eq!(y.shape(), vec![1, 4, 4, 4, 4]);
    }

    #[test]
    fn test_conv3d_with_padding() {
        let conv = Conv3d::configure(1, 2, [3, 3, 3])
            .with_padding([1, 1, 1])
            .on_device(test_device())
            .done().unwrap();
        let x = Variable::new(
            Tensor::randn(&[1, 1, 4, 4, 4], test_opts()).unwrap(), false,
        );
        let y = conv.forward(&x).unwrap();
        // Same padding: output = input size
        assert_eq!(y.shape(), vec![1, 2, 4, 4, 4]);
    }
}