OpenNN is a high-performance C++ library for advanced analytics built around neural networks. This is the API reference; for tutorials and step-by-step guides please visit opennn.net.

The library is developed by Artelnics. The source code lives at github.com/Artelnics/OpenNN.

Core classes

A typical OpenNN program revolves around five classes that work together:

Class	Role
Dataset	Data loading, partitioning, scaling, statistics
NeuralNetwork	Architecture definition and layer management
TrainingStrategy	Pairs an optimizer with a loss and runs training
ModelSelection	Hyperparameter and architecture search
TestingAnalysis	Model evaluation on held-out data

Layer types

Dense, DenseRelu, Convolutional, ConvolutionalRelu, Pooling, Pooling3d, Recurrent, Embedding, MultiHeadAttention, Normalization3d, Scaling, Unscaling, Bounding, Addition, Flatten.

Optimizers

AdaptiveMomentEstimation (Adam), StochasticGradientDescent (SGD), QuasiNewtonMethod (BFGS), LevenbergMarquardtAlgorithm.

First program

The example below loads iris.csv, builds a small MLP for regression and trains it with Adam. See ApproximationNetwork for the architecture used here.

#include "opennn/dataset.h"
#include "opennn/standard_networks.h"
#include "opennn/training_strategy.h"
 
using namespace opennn;
 
int main()
{
    // 1. Load and partition the data.
    Dataset data("iris.csv", ",", /*has_header=*/true);
 
    // 2. Build a 4-input, hidden(3), 3-output approximation MLP.
    ApproximationNetwork network({4}, {3}, {3});
 
    // 3. Pair the network with the data, choose an optimizer and a loss.
    TrainingStrategy training_strategy(&network, &data);
    training_strategy.set_optimization_algorithm("AdaptiveMomentEstimation");
    training_strategy.set_loss("MeanSquaredError");
 
    // 4. Train.
    training_strategy.train();
 
    return 0;
}

Navigating the reference

Class List — every class in alphabetical order.
Class Index — alphabetical index by name.
File List — by header file.

Resources

Tutorials: https://www.opennn.net/tutorials/
License: GNU Lesser General Public License v2.1 or later.
Issues / contributions: https://github.com/Artelnics/OpenNN