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
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203


extern crate rand;

use std::fmt;
use std::any::Any;
use std::sync::{Arc, RwLock};
use super::{
    layertype::LayerType,
    layer::Layer,
    dense::Dense,
    vectorops
};    
use super::super::{
    activation::Activation,
    neatenv::NeatEnvironment,
};    

use crate::Genome;




#[derive(Debug, Serialize, Deserialize)]
pub struct GRU {
    pub input_size: u32,
    pub memory_size: u32,
    pub output_size: u32,
    pub current_memory: Vec<f32>,
    pub current_output: Vec<f32>,
    pub f_gate: Dense,
    pub e_gate: Dense,
    pub o_gate: Dense,
}



/// implement a simple GRU layer that can be added to the neat network

impl GRU {


    pub fn new(input_size: u32, memory_size: u32, output_size: u32, act: Activation) -> Self {        
        let network_in_size = input_size + memory_size + output_size;
        GRU {
            input_size,
            memory_size,
            output_size,
            current_memory: vec![0.0; memory_size as usize],
            current_output: vec![0.0; output_size as usize],
            f_gate: Dense::new(network_in_size, memory_size, LayerType::DensePool, Activation::Sigmoid),
            e_gate: Dense::new(network_in_size, memory_size, LayerType::DensePool, Activation::Tanh),
            o_gate: Dense::new(network_in_size, output_size, LayerType::DensePool, act),
        }
    }


}




/// implement the layer trait for the GRU so it can be stored in the neat network

#[typetag::serde]
impl Layer for GRU {


    /// implement the propagation function for the GRU layer 

    #[inline]
    fn forward(&mut self, inputs: &Vec<f32>) -> Option<Vec<f32>> {
        let mut concat_input_output = self.current_output.clone();
        concat_input_output.extend(inputs);

        let mut network_input = concat_input_output.clone();
        network_input.extend(&self.current_memory);

        // calculate memory updates

        let mut forget = self.f_gate.forward(&network_input)?;
        let mut memory = self.e_gate.forward(&network_input)?;

        // figure out what to forget from the current memory

        vectorops::element_multiply(&mut self.current_memory, &forget);
        vectorops::element_invert(&mut forget);
        vectorops::element_multiply(&mut memory, &forget);
        vectorops::element_add(&mut self.current_memory, &memory);

        // add the new memory for the input to the output network of the layer

        concat_input_output.extend(&self.current_memory);

        // calculate the current output of the layer

        self.current_output = self.o_gate.forward(&concat_input_output)?;
        Some(self.current_output.clone())
    }


    fn backward(&mut self, _errors: &Vec<f32>, _learning_rate: f32) -> Option<Vec<f32>> {
        panic!("Backprop for GRU is not implemented yet");
        // let output_error = self.o_gate.backward(&errors, learning_rate)?;

        // // let delta_mem = self.current_memory

        // //     .iter()

        // //     .zip(output_error.iter())

        // //     .map(|(a, b)| {

        // //         a * b

        // //     })

        // //     .collect::<Vec<_>>();


        // // let delta_out = errors

        // //     .iter()

        // //     .zip(output_error.iter())

        // //     .map(|(a, b)| {

        // //         a * b

        // //     })

        // //     .collect::<Vec<_>>();

        
        // self.f_gate.backward(&output_error, learning_rate)?;

        // self.e_gate.backward(&output_error, learning_rate)?;


        
        // Some(errors.clone())

    }


    fn as_ref_any(&self) -> &dyn Any
        where Self: Sized + 'static
    {
        self
    }


    fn as_mut_any(&mut self) -> &mut dyn Any
        where Self: Sized + 'static
    {
        self
    }

    fn shape(&self) -> (usize, usize) {
        (self.input_size as usize, self.output_size as usize)
    }

}


/// Implement clone for the neat neural network in order to facilitate 

/// proper crossover and mutation for the network

impl Clone for GRU {

    #[inline]
    fn clone(&self) -> Self {
        GRU {
            input_size: self.input_size,
            memory_size: self.memory_size,
            output_size: self.output_size,
            current_memory: vec![0.0; self.memory_size as usize],
            current_output: vec![0.0; self.output_size as usize],
            f_gate: self.f_gate.clone(),
            o_gate: self.o_gate.clone(),
            e_gate: self.e_gate.clone(),
        }
    }
}




/// in order for the GRU layer to be evolved along with the rest of the network, Genome must be implemented 

/// so that the layer can be crossed over and measured along with other GRU layers 

impl Genome<GRU, NeatEnvironment> for GRU
    where GRU: Layer
{

    /// implement how to crossover two GRU layers 

    #[inline]
    fn crossover(child: &GRU, parent_two: &GRU, env: Arc<RwLock<NeatEnvironment>>, crossover_rate: f32) -> Option<GRU> {
        let child = GRU {
            input_size: child.input_size,
            memory_size: child.memory_size,
            output_size: child.output_size,
            current_memory: vec![0.0; child.memory_size as usize],
            current_output: vec![0.0; child.output_size as usize],
            f_gate: Dense::crossover(&child.f_gate, &parent_two.f_gate, Arc::clone(&env), crossover_rate)?,
            o_gate: Dense::crossover(&child.o_gate, &parent_two.o_gate, Arc::clone(&env), crossover_rate)?,
            e_gate: Dense::crossover(&child.e_gate, &parent_two.e_gate, Arc::clone(&env), crossover_rate)?,
        };
        Some(child)
    }


    /// get the distance between two GRU layers of the network

    #[inline]
    fn distance(one: &GRU, two: &GRU, env: Arc<RwLock<NeatEnvironment>>) -> f32 {
        let mut result = 0.0;
        result += Dense::distance(&one.f_gate, &two.f_gate, Arc::clone(&env));
        result += Dense::distance(&one.o_gate, &two.o_gate, Arc::clone(&env));
        result += Dense::distance(&one.e_gate, &two.e_gate, Arc::clone(&env));
        result
    }
}

/// implement display for the GRU layer of the network

impl fmt::Display for GRU {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        write!(f, "GRU=[input={}, memory={}, output={}]",
          self.input_size, self.memory_size, self.output_size)
    }
}