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


extern crate rand;

use rand::Rng;

use super::id::*;
use super::edge::*;
use super::activation::Activation;
use super::neurontype::NeuronType;
use super::direction::NeuronDirection;


#[derive(Deserialize, Serialize, Debug, Clone)]
pub struct NeuronLink {
    pub id: EdgeId,
    pub src: NeuronId,
    pub weight: f32,
}

impl NeuronLink {
    pub fn new(edge: &Edge) -> Self {
        Self {
            id: edge.id,
            src: edge.src,
            weight: edge.weight,
        }
    }
}

/// Neuron is a wrapper around a neuron providing only what is needed for a neuron to be added 

/// to the NEAT graph, while the neuron encapsulates the neural network logic for the specific node type,

/// Some neurons like an LSTM require more variables and different internal activation logic,

/// so encapsulating that within a normal node on the graph would be misplaced.

#[derive(Deserialize, Serialize, Debug)]
pub struct Neuron {
    pub id: NeuronId,
    outgoing: Vec<EdgeId>,
    incoming: Vec<NeuronLink>,
    activation: Activation,
    direction: NeuronDirection,
    pub neuron_type: NeuronType,
    pub activated_value: f32,
    pub deactivated_value: f32,
    pub current_state: f32,
    pub previous_state: f32,
    pub error: f32,
    pub bias: f32,
}


impl Neuron {
    pub fn new(id: NeuronId, neuron_type: NeuronType, activation: Activation, direction: NeuronDirection) -> Self {
        Neuron {
            id,
            outgoing: Vec::new(),
            incoming: Vec::new(),
            activation,
            neuron_type,
            direction,
            activated_value: 0.0,
            deactivated_value: 0.0,
            current_state: 0.0,
            previous_state: 0.0,
            error: 0.0,
            bias: rand::thread_rng().gen::<f32>(),
        }
    }

    /// Add incoming edge

    pub fn add_incoming(&mut self, edge: &Edge) {
        self.incoming.push(NeuronLink::new(edge));
    }

    /// Add outgoing edge

    pub fn add_outgoing(&mut self, edge: EdgeId) {
        self.outgoing.push(edge);
    }

    /// Update incoming edge

    pub fn update_incoming(&mut self, edge: &Edge, weight: f32) {
        if let Some(link) = self.incoming.iter_mut().find(|x| x.id == edge.id) {
            link.weight = weight;
        }
    }

    /// Remove incoming edge

    pub fn remove_incoming(&mut self, edge: &Edge) {
        self.incoming.retain(|x| x.id != edge.id);
    }

    /// Remove outgoing edge

    pub fn remove_outgoing(&mut self, edge: EdgeId) {
        self.outgoing.retain(|x| x != &edge);
    }

    /// Get incoming edge ids.

    pub fn incoming_edges(&self) -> &[NeuronLink] {
        &self.incoming
    }

    /// Get outgoing edge ids.

    pub fn outgoing_edges(&self) -> &[EdgeId] {
        &self.outgoing
    }

    /// 𝜎(Σ(w * i) + b)

    /// activate this node by calling the underlying neuron's logic for activation

    #[inline]
    pub fn activate(&mut self) {
        if self.activation != Activation::Softmax {
            match self.direction {
                NeuronDirection::Forward => {
                    self.activated_value = self.activation.activate(self.current_state);
                    self.deactivated_value = self.activation.deactivate(self.current_state);
                },
                NeuronDirection::Recurrent => {
                    self.activated_value = self.activation.activate(self.current_state + self.previous_state);
                    self.deactivated_value = self.activation.deactivate(self.current_state + self.previous_state);
                }
            }
            self.previous_state = self.current_state;
        }
    }


    /// each Neuron has a base layer of reset which needs to happen 

    /// but on top of that each neuron might need to do more internally

    #[inline]
    pub fn reset_neuron(&mut self) {
        self.error = 0.0;
        self.activated_value = 0.0;
        self.deactivated_value = 0.0;
        self.current_state = 0.0;
    }


    #[inline]
    pub fn clone_with_values(&self) -> Self {
        Neuron {
            id: self.id,
            outgoing: self.outgoing.clone(),
            incoming: self.incoming.clone(),
            current_state: self.current_state.clone(),
            previous_state: self.previous_state.clone(),
            activated_value: self.activated_value.clone(),
            deactivated_value: self.deactivated_value.clone(),
            error: self.error.clone(),
            bias: self.bias.clone(),
            activation: self.activation.clone(),
            neuron_type: self.neuron_type.clone(),
            direction: self.direction.clone()
        }
    }
}

impl Clone for Neuron {
    fn clone(&self) -> Self { 
        Neuron {
            id: self.id,
            outgoing: self.outgoing.clone(),
            incoming: self.incoming.clone(),
            current_state: 0.0,
            previous_state: 0.0,
            activated_value: 0.0,
            deactivated_value: 0.0,
            error: 0.0,
            bias: self.bias.clone(),
            activation: self.activation.clone(),
            neuron_type: self.neuron_type.clone(),
            direction: self.direction.clone()
        }
    }
}