neural_network 0.1.3

Modular neural network with an implementation of back propagation learning algorithm.
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

mod connection;
mod neuron;

use self::connection::Connection;
use self::neuron::Neuron;
use NeuralNetwork;
use rand::{ Rng, thread_rng, ThreadRng, random };
use std::f64::{ INFINITY, NEG_INFINITY };

pub struct Evolver {
    nets: Vec<(NeuralNetwork<Neuron>, f64)>,
    pub mutation_rate: f64,
    pub max_size: usize
}

macro_rules! pick {
    ($a:expr, $b:expr, $rng:ident, $mr:ident, $mid:ident) => {{
        let mut r: f64 = $rng.gen();
        if r < $mid {
            if r < $mr {
                r = $rng.gen::<f64>();
            }
            else {
                r = $a;
            }
        }
        else {
            r = $b;
        }
        r
    }}
}

impl Evolver {
    pub fn new(input_count: usize, hidden_counts: &[usize], output_count: usize,
               max_size: usize, mutation_rate: f64,
               hidden_activation: fn(f64) -> f64, output_activation: fn(f64) -> f64) -> Self {
        Evolver {
            nets: (0usize..max_size)
                .map(|_|(NeuralNetwork::new(input_count, hidden_counts, output_count, hidden_activation, output_activation), 0.0))
                .collect(),
            mutation_rate,
            max_size
        }
    }

    pub fn evolve(&mut self) {
        let mut rng = thread_rng();

        // Get the min and the max rewards
        let best_net: NeuralNetwork<Neuron>;
        let mut min_fit = INFINITY;
        {
            let mut max_index = 0usize;
            let mut max_fit = NEG_INFINITY;
            for (i, net) in self.nets.iter().enumerate() {
                if net.1 < min_fit {
                    min_fit = net.1;
                }
                if net.1 > max_fit {
                    max_fit = net.1;
                    max_index = i;
                }
            }
            best_net = self.nets[max_index].0.clone();
        }

        // Get the sum and add the min_fit to all of the
        let sum = self.nets.iter_mut()
            .fold(0.0, |acc, net| {
                net.1 += min_fit;
                acc + net.1
            });

        self.nets = (0usize..(self.max_size - 1usize)).map(|_| {
            let net_a: &NeuralNetwork<Neuron> = Evolver::pick_network(&self.nets, sum);
            let net_b: &NeuralNetwork<Neuron> = Evolver::pick_network(&self.nets, sum);
            let mid = (1.0 - self.mutation_rate) / 2.0;
            (NeuralNetwork {
                input: Evolver::mate_layers(&net_a.input, &net_b.input, self.mutation_rate, mid, &mut rng),
                hidden: net_a.hidden.iter().zip(net_b.hidden.iter())
                    .map(|(a, b)| Evolver::mate_layers(a, b, self.mutation_rate, mid, &mut rng))
                    .collect(),
                output: Evolver::mate_layers(&net_a.output, &net_b.output, self.mutation_rate, mid, &mut rng),
                hidden_activation: net_a.hidden_activation,
                output_activation: net_a.output_activation
            }, 0.0)
        }).collect();
        self.nets.push((best_net, 0.0));
    }
    fn mate_layers(layer_a: &Vec<Neuron>, layer_b: &Vec<Neuron>, mutation_rate: f64, mid: f64, rng: &mut ThreadRng) -> Vec<Neuron> {
        layer_a.iter().zip(layer_b.iter())
            .map(|(n_a, n_b)|
                Neuron {
                    bias: pick!(n_a.bias, n_b.bias, rng, mutation_rate, mid),
                    output: 0.0,
                    connections: n_a.connections.iter().zip(n_b.connections.iter())
                        .map(|(c_a, c_b)| Connection {
                            weight: pick!(c_a.weight, c_b.weight, rng, mutation_rate, mid)
                        })
                        .collect()
                }
            ).collect()
    }
    fn pick_network<'a>(nets: &'a Vec<(NeuralNetwork<Neuron>, f64)>, sum: f64) -> &'a NeuralNetwork<Neuron> {
        let mut sum = sum;
        let rand: f64 = random();
        for net in nets.iter() {
            sum -= net.1;
            if sum <= rand {
                return &net.0;
            }
        }
        &nets.first().unwrap().0
    }

    pub fn pulse(&mut self, index: usize, input: &[f64]) -> Vec<f64> {
        self.nets[index].0.pulse(input)
    }

    pub fn get_fitness(&self, index: usize) -> f64 {
        self.nets[index].1
    }
    pub fn set_fitness(&mut self, index: usize, x: f64) {
        self.nets[index].1 = x;
    }
    pub fn increment_fitness(&mut self, index: usize, x: f64) {
        self.nets[index].1 += x;
    }

    pub fn len(&self) -> usize {
        self.nets.len()
    }
}