Custom-Neural-network/main.cpp at main · Aldric-L/Custom-Neural-network · GitHub

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
#include "AKML.hpp"
#include "NeuralFunctions.hpp"
#include "GeneticAlgorithm.hpp"

#include <iostream>
#include <chrono>

int main (){
    using SINGLETON = std::array <std::array <float, 1>, 1>;

    std::cout << "Hello, terrible world!" << std::endl;

    akml::NeuralNetwork *bestnet;

    // XOR DIM 2
    //akml::NeuralNetwork::initialize_list_type xor2_initList = {{ std::make_pair(2, nullptr), std::make_pair(2, &akml::ActivationFunctions::RELU), std::make_pair(1, &akml::ActivationFunctions::SIGMOID) }};
    /*akml::NeuralNetwork::initialize_list_type xor2_initList = {{ std::make_pair(2, nullptr), std::make_pair(2, &akml::ActivationFunctions::SIGMOID), std::make_pair(1, &akml::ActivationFunctions::SIGMOID) }};
    std::vector<akml::DynamicMatrix <float>> inputs = {{ akml::make_dynamic_vector<float>(1.f,0.f), akml::make_dynamic_vector<float>(0.f,0.f), akml::make_dynamic_vector<float>(0.f,1.f), akml::make_dynamic_vector<float>(1.f,1.f) }};

    std::vector<akml::DynamicMatrix <float>> outputs = {{ (SINGLETON){{ {1} }}, (SINGLETON){{ {0} }}, (SINGLETON){{ {1} }}, (SINGLETON){{ {0} }} }};*/

    /*akml::GeneticAlgorithm ga (100, xor2_initList);
    bestnet = ga.trainNetworks(5000, inputs, outputs);*/

    // XOR DIM 3
    akml::NeuralNetwork::initialize_list_type xor3_initList = {{ std::make_pair(3, nullptr), std::make_pair(3, &akml::ActivationFunctions::RELU), std::make_pair(3, &akml::ActivationFunctions::RELU), std::make_pair(1, &akml::ActivationFunctions::SIGMOID) }};

    std::vector<akml::DynamicMatrix <float>> inputs = {{
        akml::make_dynamic_vector<float>(1.f,0.f,0.f), akml::make_dynamic_vector<float>(0.f,0.f, 1.f),
        akml::make_dynamic_vector<float>(0.f,1.f,0.f), akml::make_dynamic_vector<float>(1.f,1.f,0.f),
        akml::make_dynamic_vector<float>(1.f,0.f,1.f), akml::make_dynamic_vector<float>(0.f,1.f,1.f),
        akml::make_dynamic_vector<float>(1.f,1.f,1.f) }};

    std::vector<akml::DynamicMatrix <float>> outputs = {{ (SINGLETON){{ {1} }}, (SINGLETON){{ {1} }},
        (SINGLETON){{ {1} }}, (SINGLETON){{ {0} }},
        (SINGLETON){{ {0} }}, (SINGLETON){{ {0} }}, (SINGLETON){{ {0} }} }};

    bestnet = new akml::NeuralNetwork (xor3_initList);

    auto start = std::chrono::high_resolution_clock::now();
    // Genetic algorithm
    /*akml::GeneticAlgorithm ga (100, xor3_initList);
    bestnet = ga.trainNetworks(5000, inputs, outputs);*/

    // Gradient descent
    bestnet = new akml::NeuralNetwork (xor3_initList);
    bestnet->adamGradientTraining(inputs, outputs, 3, 5000);

    auto end = std::chrono::high_resolution_clock::now();
    std::chrono::duration<double> duration = end - start;
    std::cout << "\nExecutionTime=" << duration.count() << "\n";

    for (std::size_t inputid(0); inputid<inputs.size(); inputid++){
        std::cout << "\nTesting with " << std::endl;
        std::cout << inputs[inputid];
        std::cout << "Output :" << std::endl;
        std::cout << bestnet->process(inputs[inputid]);
        std::cout << "Output expected :" << std::endl;
        std::cout << outputs[inputid];
    }

    return 0;
}