nn-cpp/main.cpp at main · at-src/nn-cpp · 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
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
#include <iostream>
#include <vector>
#include <iomanip>
#include "network.cpp"

int main() {
    std::cout << "neural network demo" << std::endl << std::endl;

    std::cout << "=== test 1: xor classification ===" << std::endl;
    {
        NeuralNet model;
        model.add_layer(new DenseLayer(2, 4));
        model.add_layer(new ReLULayer());
        model.add_layer(new DenseLayer(4, 1));

        std::vector<std::vector<double>> X = {{0, 0}, {0, 1}, {1, 0}, {1, 1}};
        std::vector<std::vector<double>> y = {{0}, {1}, {1}, {0}};

        for (int epoch = 0; epoch < 5000; epoch++) {
            double total_loss = 0.0;
            for (size_t i = 0; i < X.size(); i++) {
                auto logits = model.predict(X[i]);
                auto preds = apply_sigmoid(logits);
                Loss loss = compute_bce(preds, y[i]);
                total_loss += loss.value;
                model.backward_pass(loss.gradient, 0.1);
            }
            if (epoch % 1000 == 0) {
                std::cout << "epoch " << epoch << " loss " << total_loss / X.size() << std::endl;
            }
        }

        std::cout << std::fixed << std::setprecision(6);
        for (const auto& input : X) {
            auto logit = model.predict(input);
            auto output = apply_sigmoid(logit);
            std::cout << (int)input[0] << " " << (int)input[1] << " -> " << output[0] << std::endl;
        }
    }

    std::cout << std::endl << "=== test 2: and gate ===" << std::endl;
    {
        NeuralNet model;
        model.add_layer(new DenseLayer(2, 3));
        model.add_layer(new TanhLayer());
        model.add_layer(new DenseLayer(3, 1));

        std::vector<std::vector<double>> X = {{0, 0}, {0, 1}, {1, 0}, {1, 1}};
        std::vector<std::vector<double>> y = {{0}, {0}, {0}, {1}};

        for (int epoch = 0; epoch < 3000; epoch++) {
            double total = 0.0;
            for (size_t i = 0; i < X.size(); i++) {
                auto out = model.predict(X[i]);
                auto sig = apply_sigmoid(out);
                Loss l = compute_bce(sig, y[i]);
                total += l.value;
                model.backward_pass(l.gradient, 0.1);
            }
            if (epoch % 1000 == 0) {
                std::cout << "epoch " << epoch << " loss " << total / X.size() << std::endl;
            }
        }

        std::cout << std::fixed << std::setprecision(6);
        for (const auto& input : X) {
            auto out = model.predict(input);
            auto sig = apply_sigmoid(out);
            std::cout << (int)input[0] << " " << (int)input[1] << " -> " << sig[0] << std::endl;
        }
    }

    std::cout << std::endl << "=== test 3: or gate ===" << std::endl;
    {
        NeuralNet model;
        model.add_layer(new DenseLayer(2, 2));
        model.add_layer(new LeakyReLULayer(0.01));
        model.add_layer(new DenseLayer(2, 1));

        std::vector<std::vector<double>> X = {{0, 0}, {0, 1}, {1, 0}, {1, 1}};
        std::vector<std::vector<double>> y = {{0}, {1}, {1}, {1}};

        for (int epoch = 0; epoch < 2000; epoch++) {
            double total = 0.0;
            for (size_t i = 0; i < X.size(); i++) {
                auto out = model.predict(X[i]);
                auto sig = apply_sigmoid(out);
                Loss l = compute_bce(sig, y[i]);
                total += l.value;
                model.backward_pass(l.gradient, 0.15);
            }
            if (epoch % 500 == 0) {
                std::cout << "epoch " << epoch << " loss " << total / X.size() << std::endl;
            }
        }

        std::cout << std::fixed << std::setprecision(6);
        for (const auto& input : X) {
            auto out = model.predict(input);
            auto sig = apply_sigmoid(out);
            std::cout << (int)input[0] << " " << (int)input[1] << " -> " << sig[0] << std::endl;
        }
    }

    return 0;
}