triton_grow 2.2.7

A self sustaining growing neural net that can repair itself until reaching a desired accuracy
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

use serde::{Serialize, Deserialize};
use crate::network::{matrix::Matrix,activations::Activations, input::Input, matrix3d::Matrix3D};
use super::{layers::Layer, distributions::Distributions};

#[derive(Serialize, Deserialize)]
pub struct Convolutional{
    filter_weights: Matrix3D,
    filter_biases: Vec<f32>,
    data: Matrix3D,
    stride: usize,
    filters: usize,
    shape: (usize, usize),
    input_shape: (usize, usize, usize),
    output_shape: (usize, usize, usize),
    loss: f32,
    
    beta1: f32,
    beta2: f32,
    epsilon: f32,
    time: usize,

    m_weights: Matrix3D,
    v_weights: Matrix3D,
    m_biases: Vec<f32>,
    v_biases: Vec<f32>,

    activation_fn: Activations,
    learning_rate: f32
}

impl Convolutional{
    pub fn new(filters: usize, kernel_size: (usize, usize), input_shape: (usize, usize, usize), stride: usize, activation_fn: Activations, learning_rate: f32, seed: &Option<String>, input_size: usize) -> Convolutional {
        let distribution = match activation_fn {
            Activations::TANH | Activations::SIGMOID | Activations::SOFTMAX => Distributions::Xavier(input_size, kernel_size.0 * kernel_size.1),
            Activations::RELU | Activations::LEAKYRELU => Distributions::He(input_size)
        };
        let mut res = Convolutional{
            filter_weights: Matrix3D::new_random(kernel_size.0, kernel_size.1, filters, seed, &distribution),
            m_weights: Matrix3D::new_empty(kernel_size.0, kernel_size.1, filters),
            v_weights: Matrix3D::new_empty(kernel_size.0, kernel_size.1, filters),

            filter_biases: Matrix::new_random(1, filters, seed, &distribution).to_param(),
            m_biases: vec![0.0; filters],
            v_biases: vec![0.0; filters],

            data: Matrix3D::new_empty(0,0,0),
            stride,
            filters,
            input_shape,
            shape: kernel_size,
            activation_fn,
            learning_rate,
            output_shape: (0,0,0),
            loss: 1.0,

            beta1: 0.0,
            beta2: 0.0,
            epsilon: 0.0,
            time: 1
        };

        let res_len = Convolutional::get_res_size(input_shape.0, kernel_size.0, 0, stride);
        let res_width = Convolutional::get_res_size(input_shape.1, kernel_size.1, 0, stride);
        
        res.data = Matrix3D::new_empty(res_len, res_width, filters);
        res.output_shape = (res_len, res_width, 1);

        (res.beta1, res.beta2) = res.get_betas();
        res.epsilon = res.get_epsilon();

        res
    }
    fn get_betas(&self) -> (f32, f32) {
        (0.9, 0.999)
    }
    fn get_epsilon(&self) -> f32{
        1e-10
    }
    pub fn convolute(&mut self, idx: usize, input: Matrix) {
        let kernel = self.filter_weights.get_slice(idx);
        let mut output = Matrix::new_empty(self.output_shape.0, self.output_shape.1);

        let mut x: usize;
        let mut y: usize = 0;

        for output_x in 0..output.columns {
            x = 0;
            for output_y in 0..output.rows {
                let sum = input.get_sub_matrix(x, y, kernel.rows, kernel.columns).dot_multiply(&kernel).sum();
                output.data[output_y][output_x] = sum;

                x += self.stride;
            }
            y += 1;
        }

        //println!("{}", output);
        self.data.set_slice(idx, output);
    }
    fn get_res_size(w: usize, k: usize, p: usize, s:usize) -> usize {
        (w - k + 2*p) / s + 1
    }
}

#[typetag::serde]
impl Layer for Convolutional {
    fn forward(&mut self,inputs: &Box<dyn Input>) -> Box<dyn Input> {
        let input_mat = Matrix3D::from(inputs.to_param_3d());
        for i in 0..input_mat.layers {
            for j in 0..self.filters {
                self.convolute(j, input_mat.get_slice(i));
            }
        }
        self.data = self.data.clone() + &self.filter_biases;
        Box::new(self.data.clone())
    }
    fn backward(&mut self,gradients:Box<dyn Input> ,errors:Box<dyn Input> ,data:Box<dyn Input>) -> Box<dyn Input> {
        let mut gradients_mat = Matrix3D::from(gradients.to_param_3d());
        let mut errors_mat = Matrix3D::from(errors.to_param_3d());
        let data_mat = Matrix3D::from(data.to_param_3d());

        gradients_mat = gradients_mat.dot_multiply(&errors_mat) * self.learning_rate;
        errors_mat = self.filter_weights.clone().transpose() * &errors_mat;

        self.loss = 0.0;
        errors_mat.to_param().iter().for_each(|error| {
            self.loss += error.powi(2);
        });
        self.loss = self.loss / errors_mat.to_param().len() as f32;

        panic!();
    }
    fn get_data(&self) -> Box<dyn Input> {
        self.data.to_box()
    }
    fn shape(&self) -> (usize,usize,usize) {
        self.input_shape
    }
    fn get_loss(&self) -> f32 {
        self.loss
    }
    fn update_gradient(&self) -> Box<dyn Input> {
        //TODO
        self.data.to_box()
    }
}