Neural Network MLP (C++)

A Multi-Layer Perceptron neural network implemented from scratch in C++ with zero external dependencies. Available as both static and dynamic libraries.

Features

• Pure C++ implementation with no ML framework dependencies
• Configurable layer count and neuron count
• Bias neurons in each layer (except output)
• ReLU activation for hidden layers
• Weight modification for training
• Network truncation support (weight averaging)
• Available as static (.a) and dynamic (.so) libraries

Building

./build.sh

This creates both lib/libredeneural.a and lib/libredeneural.so.

Usage

Include in Your Project

Copy include/redeNeural.hpp to your project's include path and link against the library.

CMake Integration

# Static
target_link_libraries(your_target /path/to/lib/libredeneural.a)

# Dynamic
target_link_libraries(your_target /path/to/lib/libredeneural.so)

Example

#include <redeNeural.hpp>
#include <vector>

int main() {
    // 3 input neurons, 2 hidden layers, 2 output neurons
    RedeNeural network(3, 2, 2);

    std::vector<double> inputs = {0.5, -0.5, 1.0};
    std::vector<bool> output = network.iniciar(inputs);

    return 0;
}

Architecture

• Input layer with configurable neuron count + bias
• N hidden layers with ReLU activation + bias
• Output layer returning boolean vector
• Weight initialization and modification API for training

Requirements

• C++17
• No external dependencies

License

MIT