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


use std::collections::HashMap;

use super::id::*;


/// Tracer keeps track of historical metadata for neurons to keep track

/// of their activated values and derivatives so backpropagation (through time)

/// is available for batch processing and weight updates

#[derive(Deserialize, Serialize, Debug, Clone)]
pub struct Tracer {
    pub neuron_activation: HashMap<NeuronId, Vec<f32>>,
    pub neuron_derivative: HashMap<NeuronId, Vec<f32>>,
    pub max_neuron_index: usize,
    pub index: usize,
}



impl Tracer {

    pub fn new() -> Self {
        Tracer {
            neuron_activation: HashMap::new(),
            neuron_derivative: HashMap::new(),        
            max_neuron_index: 0,
            index: 0,
        }
    }



    /// reset the tracer. The backprop works off of indexed values so when the

    /// layer is reset, the tracer must be reset as well

    pub fn reset(&mut self) {
        self.neuron_activation = HashMap::new();
        self.neuron_derivative = HashMap::new();
        self.index = 0;        
    }



    /// update a neuron and add it's activated value 𝜎(Σ(w * i) + b)

    pub fn update_neuron_activation(&mut self, neuron_id: &NeuronId, neuron_value: f32) {
        if self.neuron_activation.contains_key(&neuron_id) {
            let states = self.neuron_activation.get_mut(&neuron_id).unwrap();
            states.push(neuron_value);

            // keep track of how many values are being kept track of so the list's don't 

            // have to resize after one iteration, speeds things up as time goes on 

            if states.len() > self.max_neuron_index {
                self.max_neuron_index += 1;
            }
        } else {
            let mut temp = Vec::with_capacity(self.max_neuron_index);
            temp.push(neuron_value);
            self.neuron_activation.insert(*neuron_id, temp);
        }
    }



    /// update a neuron and add it's derivative of it's activated value to the tracer

    pub fn update_neuron_derivative(&mut self, neuron_id: &NeuronId, neuron_d: f32) {
        if self.neuron_derivative.contains_key(&neuron_id) {
            let states = self.neuron_derivative.get_mut(&neuron_id).unwrap();
            states.push(neuron_d);
        } else {
            let mut temp = Vec::with_capacity(self.max_neuron_index);
            temp.push(neuron_d);
            self.neuron_derivative.insert(*neuron_id, temp);
        }
    }



    /// return the activated value of a neuron at the current index 

    pub fn neuron_activation(&self, neuron_id: NeuronId) -> f32 {
        if !self.neuron_activation.contains_key(&neuron_id) {
            panic!("Tracer neuron state doesn't contain uuid: {:?}", neuron_id);
        }
        self.neuron_activation.get(&neuron_id).unwrap()[self.index - 1]
    }



    /// return the derivative of a neuron at the current index 

    pub fn neuron_derivative(&self, neuron_id: NeuronId) -> f32 {
        if !self.neuron_derivative.contains_key(&neuron_id) {
            panic!("Tracer neuron state doesn't contain uuid: {:?}", neuron_id);
        }
        self.neuron_derivative.get(&neuron_id).unwrap()[self.index - 1]
    }



}