jiro_nn 0.8.1

Neural Networks framework with model building & data preprocessing features.
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

use crate::{
    layer::LearnableLayer,
    linalg::{Matrix, MatrixTrait, Scalar},
    vision::{
        conv_initializers::ConvInitializers, conv_optimizer::ConvOptimizers, image::Image,
        image::ImageTrait,
    },
};

use crate::vision::image_layer::ImageLayer;

use super::ConvLayer;

#[derive(Debug)]
pub struct DirectConvLayer {
    pub kernels: Image,
    biases: Image,
    input: Option<Image>,
    kernels_optimizer: ConvOptimizers,
    biases_optimizer: ConvOptimizers,
}

impl DirectConvLayer {
    pub fn new(
        krows: usize,
        kcols: usize,
        in_chans: usize,
        kernels_initializer: ConvInitializers,
        biases_initializer: ConvInitializers,
        kernels_optimizer: ConvOptimizers,
        biases_optimizer: ConvOptimizers,
    ) -> Self {
        Self {
            kernels: kernels_initializer.gen_image(krows, kcols, in_chans, 1),
            biases: biases_initializer.gen_image(1, 1, in_chans, 1),
            input: None,
            kernels_optimizer,
            biases_optimizer,
        }
    }

    pub fn out_img_dims_and_channels(
        in_rows: usize,
        in_cols: usize,
        in_chans: usize,
        krows: usize,
        kcols: usize,
    ) -> (usize, usize, usize) {
        let out_rows = in_rows - krows + 1;
        let out_cols = in_cols - kcols + 1;
        (out_rows, out_cols, in_chans)
    }
}

impl ImageLayer for DirectConvLayer {
    fn forward(&mut self, input: Image) -> Image {
        let mut channels = vec![];
        for c in 0..input.channels() {
            let channel = input.get_channel_across_samples(c);

            let kernel = self.kernels.get_channel(c);

            let correlated = channel.cross_correlate(&kernel);

            if self.biases.image_dims().0 != correlated.image_dims().0 {
                self.biases =
                    self.biases
                        .tile(correlated.image_dims().0, correlated.image_dims().1, 1, 1);
            }

            let bias = self.biases.get_channel(c);

            let result_channel = correlated.component_add(&bias);
            channels.push(result_channel);
        }
        let res = Image::join_channels(channels);
        self.input = Some(input);
        res
    }

    fn backward(&mut self, epoch: usize, output_gradient: Image) -> Image {
        let input = self.input.as_ref().unwrap();

        let mut input_grad_channels = vec![];
        for i in 0..input.channels() {
            let kernel = self.kernels.get_channel(i);
            let output_grad_i = output_gradient.get_channel_across_samples(i);
            let correlated = output_grad_i.convolve_full(&kernel);
            input_grad_channels.push(correlated);
        }
        let input_grad = Image::join_channels(input_grad_channels);

        let mut kern_grad_channels = vec![];
        for i in 0..input.channels() {
            let output_grad_i = output_gradient.get_channel_across_samples(i).sum_samples();
            let input_i = input.get_channel_across_samples(i);
            let correlated = input_i.cross_correlate(&output_grad_i).sum_samples();
            kern_grad_channels.push(correlated);
        }
        let kern_grad = Image::join_channels(kern_grad_channels);

        let mut biases_grad_channels = vec![];
        for c in 0..self.biases.channels() {
            let channel = output_gradient.get_channel_across_samples(c);
            let channel = channel.sum_samples();
            biases_grad_channels.push(channel);
        }
        let biases_grad = Image::join_channels(biases_grad_channels);

        self.kernels = self
            .kernels_optimizer
            .update_parameters(epoch, &self.kernels, &kern_grad);
        self.biases = self
            .biases_optimizer
            .update_parameters(epoch, &self.biases, &biases_grad);
        input_grad
    }
}

impl LearnableLayer for DirectConvLayer {
    fn get_learnable_parameters(&self) -> Vec<Vec<Scalar>> {
        let mut params = self.kernels.flatten().get_data_col_leading();
        params.push(self.biases.flatten().get_column(0));
        params
    }

    fn set_learnable_parameters(&mut self, params_matrix: &Vec<Vec<Scalar>>) {
        let mut kernels = params_matrix.clone();
        let biases = kernels.pop().unwrap();
        self.kernels = Image::from_samples(
            &Matrix::from_column_leading_vector2(&kernels),
            self.kernels.channels(),
        );
        self.biases =
            Image::from_samples(&Matrix::from_column_vector(&biases), self.biases.channels());
    }
}

impl ConvLayer for DirectConvLayer {
    fn scale_kernels(&mut self, scale: Scalar) {
        self.kernels = self.kernels.scalar_mul(scale);
    }
}