Overview

This program initializes and trains a deep neural network using predefined training data. It consists of an input layer, a hidden layer, and an output layer. The trained model is then used to make predictions.

Code Explanation

Initialize Network Architecture

LayerLayout layout;
layout.qtyInputs = 23;
layout.qtyOutputs = 10;
layout.qtyHiddenLayers = {18}; // e.g. {10, 13, 18} for 3 hidden layer with 10, 13 and 18 neurons

DeepLayers dl(layout);
std::cout << dl.TotalLogicLayers; // print logical layers
std::cout << dl.TotalNeurons;     // print total neurons of network

NeuralNetwork::DeepNetwork DDN(dl);

The neural network is initialized with:

• 23 input neurons
• 10 output neurons
• 1 hidden layer with 18 neurons

Define Training Data

std::vector<NeuralNetwork::Dataset> traindata = {
    { .input = {1, 1, 1, ...}, .target = {1, 0, 0, ...} },
    { .input = {0, 1, 0, ...}, .target = {0, 1, 0, ...} },
    ...
};

The dataset consists of multiple input vectors, each paired with a corresponding target output vector.

Train the Network

DDN.Train(traindata, 100000);

The neural network is trained using the dataset for 100,000 iterations.

Make Predictions

for (const NeuralNetwork::Dataset& data : traindata) {
    std::vector<double> res = DDN.GetOutput(data.input);
}

The trained model generates predictions for each input in the training dataset.

Test with a Custom Input

std::vector<double> res = DDN.GetOutput(std::vector<double>{
    1, 1, 1,
    1, 0, 1, 
    1, 1, 1,
    1, 0, 1,
    1, 1, 1,
});

An independent test input is provided, and the resulting output is printed to the console.

Output Results

for (double r : res) {
    std::cout << r << std::endl;
}

The program prints the predicted output values to the console.