libdt 2.0.3

Library for training neural networks.
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

use nalgebra as na;
use std::cmp::Ordering;

use na::DMatrix;
use na::DVector;
use na::Matrix;
use na::Vector;
use na::base::dimension as dim;
use na::base::VecStorage;

use super::Layer;

pub struct SoftMaxLayer<const SIZE: usize> {
    pub signal: Vector::<f64, dim::Dyn,
                         VecStorage::<f64, dim::Dyn, dim::U1>>,
    chain_element: Matrix::<f64, dim::Dyn, dim::Dyn,
                            VecStorage::<f64, dim::Dyn, dim::Dyn>>,
}

impl<const SIZE: usize> SoftMaxLayer<SIZE>
{
    pub fn new() -> Self {
        Self {
            signal: Vector::from_element_generic(dim::Dyn(Self::NEURONS_OUT),
                                                 dim::U1, 0f64),
            chain_element: Matrix::from_element_generic(dim::Dyn(Self::NEURONS_OUT),
                                                        dim::Dyn(Self::NEURONS_IN), 0f64),
        }
    }
}

fn softmax_d(signal: &DVector<f64>, i: usize, j: usize) -> f64 {
    (if i == j { signal[i] } else { 0f64 }) -
    signal[i] * signal[j]
}

impl<const SIZE: usize> Layer for SoftMaxLayer<SIZE> {
    const PARAMS_CNT: usize = 0;
    const NEURONS_IN: usize = SIZE;
    const NEURONS_OUT: usize = SIZE;

    unsafe fn eval_unchecked(p: &[f64], x: DVector<f64>) -> DVector<f64> {
        Self::eval(p, x)
    }

    fn eval(_p: &[f64], x: DVector<f64>) -> DVector<f64> {
        assert_eq!(x.len(), Self::NEURONS_IN);

        let mut x = x;

        let mut max_elem: f64 = 0f64;
        for xi in x.iter_mut() {
            match (*xi).partial_cmp(&max_elem) {
                Some(Ordering::Greater) => { max_elem = *xi; }
                _ => {}
            }
        }

        let mut layer_sum: f64 = 0f64;
        for xi in x.iter_mut() {
            *xi = (*xi - max_elem).exp();
            layer_sum += *xi;
        }

        for xi in x.iter_mut() {
            *xi /= layer_sum;
        }

        x
    }

    fn forward(&mut self, p: &[f64], x: DVector<f64>) -> DVector<f64> {
        assert_eq!(p.len(), Self::PARAMS_CNT);
        assert_eq!(x.len(), Self::NEURONS_IN);

        self.signal = Self::eval(p, x);
        self.signal.clone()
    }

    fn backward(&mut self, _p: &[f64]) {
        self.chain_element = DMatrix::from_element_generic(
            dim::Dyn(Self::NEURONS_OUT), dim::Dyn(Self::NEURONS_IN), 0f64);
        for i in 0..SIZE {
            for j in 0..SIZE {
                self.chain_element[(i, j)] = softmax_d(&self.signal, i, j);
            }
        }
    }

    fn chain_element(&self) -> &DMatrix<f64> {
        &self.chain_element
    }

    fn chain_end(&self, _x: &DVector<f64>) -> DMatrix<f64>
    {
        DMatrix::from_element_generic(
            dim::Dyn(Self::NEURONS_OUT),
            dim::Dyn(Self::PARAMS_CNT), 0f64)
    }

    fn default_initial_params() -> Vec<f64> {
        Vec::new()
    }
}

#[cfg(test)]
mod tests {
    use float_eq::assert_float_eq;
    use super::*;

    #[test]
    fn test_eval() {
        let x = DVector::from_column_slice(
            na::vector![1f64, 2f64].as_slice());
        
        let y = SoftMaxLayer::<2>::eval(&[], x);
        assert_eq!(y.len(), 2);
        assert_float_eq!(y[0], 0.2689414214f64, abs <= 0.000_000_000_1);
        assert_float_eq!(y[1], 0.7310585786f64, abs <= 0.000_000_000_1);
    }

    #[test]
    fn test_forward() {
        let x = DVector::from_column_slice(
            na::vector![2f64, -3f64].as_slice());
        let mut layer = SoftMaxLayer::<2>::new();
        
        let y = layer.forward(&[], x);
        assert_eq!(y.len(), 2);
        assert_float_eq!(y[0], 0.9933071491f64, abs <= 0.000_000_000_1);
        assert_float_eq!(y[1], 0.006692850924f64, abs <= 0.000_000_000_1);
    }

    #[test]
    fn test_backward() {
        let x = DVector::from_column_slice(
            na::vector![1f64, 2f64].as_slice());
        let mut layer = SoftMaxLayer::<2>::new();
        
        let _ = layer.forward(&[], x);
        layer.backward(&[]);

        let chain_element = layer.chain_element;
        assert_eq!(chain_element.nrows(), 2);
        assert_eq!(chain_element.ncols(), 2);
        assert_float_eq!(chain_element[(0, 0)], 0.2689414214f64*(1f64 - 0.2689414214f64), abs <= 0.000_000_000_1);
        assert_float_eq!(chain_element[(0, 1)], -0.2689414214f64*0.7310585786f64, abs <= 0.000_000_000_1);
        assert_float_eq!(chain_element[(1, 0)], -0.7310585786f64*0.2689414214f64, abs <= 0.000_000_000_1);
        assert_float_eq!(chain_element[(1, 1)], 0.7310585786f64*(1f64 - 0.7310585786f64), abs <= 0.000_000_000_1);
    }

    #[test]
    fn test_chain_end() {
        let x = DVector::from_column_slice(
            na::vector![1f64, 2f64].as_slice());
        let layer = SoftMaxLayer::<2>::new();

        assert_eq!(layer.chain_end(&x).ncols(), 0);
    }
}