FlexNN

Fully connected neural network built from scratch with flexible n-layer design and multiple activations.

Example Usage

Below is a minimal example of how to use the FlexNN NeuralNetwork library in your own C++ project:

#include <Eigen/Dense>
#include "FlexNN.h"
 
int main() {
    // Prepare your data (X: features, Y: labels)
    Eigen::MatrixXd X; // shape: (num_samples, num_features)
    Eigen::VectorXd Y; // shape: (num_samples)
    FlexNN::readCSV_XY("data/mnist-digit-recognition.csv", X, Y);
 
    // Normalize input data if needed
    X = X.array() / 255.0;
 
    // (Optional) Split data into training and test sets
    auto splits = FlexNN::splitXY(X, Y, {0.8, 0.2});
    Eigen::MatrixXd X_train = splits[0].first;
    Eigen::VectorXd Y_train = splits[0].second;
    Eigen::MatrixXd X_test = splits[1].first;
    Eigen::VectorXd Y_test = splits[1].second;
 
    // Transpose for (features, samples) shape expected by the network
    X_train.transposeInPlace();
    X_test.transposeInPlace();
 
    // Create a neural network with desired layers
    FlexNN::NeuralNetwork nn({
        FlexNN::Layer(X_train.rows(), 64, "relu"),
        FlexNN::Layer(64, 10, "softmax")
    });
 
    // Train the network
    nn.train(X_train, Y_train, 0.5, 100);
 
    // Evaluate accuracy
    double train_acc = nn.accuracy(X_train, Y_train);
    double test_acc = nn.accuracy(X_test, Y_test);
 
    std::cout << "Training accuracy: " << train_acc * 100 << "%" << std::endl;
    std::cout << "Test accuracy: " << test_acc * 100 << "%" << std::endl;
 
    return 0;
}

Note:

• You can customize the network architecture by changing the number and type of layers.
• Make sure your data is in the correct format and normalized as needed.
• See the src/main.cpp file for a more complete example.

Requirements

The C++ Neural Network library uses Eigen3 for matrix operations and OpenMP for multithreading. The project uses CMake for building the code.

On Ubuntu, you can install these requirements using:

sudo apt install cmake libeigen3-dev

Building and Running the Project

To build the project, follow these steps:

1. Open a terminal and navigate to the project directory.
2. Create a build directory:

   mkdir build
   cd build

3. Run CMake to configure the project:

   cmake ..

4. Build the project:

   cmake --build .

5. To run the program, go into the project root directory (so that dataset path is correctly resolved)

   cd ..

   And then run the main executable present inside the build folder by:

   ./build/main

API Reference

For details on the code structure, available classes, and how to use FlexNN in your own projects, please visit the full documentation here: https://docs.nalinangrish.me/FlexNN.

Known Issues / Limitations

• Only CPU computation is supported (no GPU).
• No support for convolutional or recurrent layers.
• Training large models may be slow.
• Currently there is no support for saving model architecture and weights, so you need to train the model on every execution.

License

This project is licensed under the Apache-2.0 License. See the LICENSE file for more details.