rusih100 / simple-neural-network

🧬 A simple scalable evolutionary neural network in C++

simple_neural_network

🧬 A simple scalable evolutionary neural network in C++

Description

Hi. This is a simple neural network model that is trained by evolution.
The neural network is scalable, that is, the number of input and output neurons is set, as well as the number of hidden layers with the same number of neurons on each layer is set. The sigmoid function is used as the activation function.
A neural network consists of two main classes Network_training and Neuron_network.

Example of creating a neural network

In this example, we will create a neural network that performs the XOR operation.

x	y	XOR
0	0	0
0	1	1
1	0	1
1	1	0

Let's define datasets and expected results for these datasets.
The neural network takes a vector of values as input and returns the result as a vector.

vector <vector<float>> dataset_input = {
            {0, 0}, {0, 1}, {1, 0}, {1, 1}
    };

vector <vector<float>> dataset_output = {
            {0}, {1}, {1}, {0}
    };

Create an instance of the Neuron_network class with the parameters:
2 hidden layers (hidden_layers_n), 3 neurons on each hidden layer (neurons_per_layer_n), 2 input neurons (inputs_n), 1 output neuron (outputs_n).

Neuron_network XOR(2, 3, 2, 1);

Let's train a neural network with these parameters for 100,000 epochs.
To do this, create an instance of class Network_training and pass it our neural network and datasets.
The method training() will return us a trained neural network

Network_training Training_network(dataset_input, dataset_output, XOR);  
XOR = Training_network.training(100000);

Let's test the work of a neural network on a data set:

cout << "Datasets result: \n";
    for (auto & set : dataset_input) {
        cout << "[ ";
        for (float i : set) {
            cout << i << " ";
        }
        cout << "] - ";
        cout << XOR.run(set)[0] << "\n";
    }

Results:

Datasets result:
[ 0 0 ] - 1.13952e-17
[ 0 1 ] - 1
[ 1 0 ] - 1
[ 1 1 ] - 9.37329e-10

The weights of the neural network can be viewed using the method print_weights():

1) Weights:
-27.697 8.582 -16.824
8.101 -27.766 -15.757

2) Weights:
-7.891 -5.387 6.158
23.374 -7.467 9.219
7.007 14.167 -26.493

3) Weights:
-0.27
-58.941
17.635

simple-neural-network's Issues

Скорее всего утечка пямяти

Я использую твою нейронную сеть в своём проекте, и у меня произошла утечка памяти. Visual Studio указывает на float**, они постоянно создаются, возможно где-то это моя ошибка, но это как задачка для тебя, я буду рад если ты попробуешь исправить это.

EDIT 1: Замечаю что если я не запускаю нейронную сеть утечки нет.
EDIT 2:
for (auto & layer : network_layers) {
temp = layer.run(temp);
}
Утечка происходит здесь.

EDIT 3:

Я отыскал ошибку. Оператор присваивания у матрицы не удалял прошлые данные, из-за этого происходила утечка