Straightforward Neural Network

Straightforward Neural Network is an open source neural network library in C++20 optimized for CPU. The goal of this library is to make the use of neural networks as easy as possible.

Documentation

See the full documentation here.

Classification datasets results

Dataset Name	Data type	Problem type	Score	Number of Parameters
Audio Cats and Dogs	audio	classification	91.04% Accurracy	382
Daily min temperatures	time series	regression	1.42 Mean Absolute Error	30
CIFAR-10	image	classification	61.77% Accurracy	207210
Fashion-MNIST	image	classification	89.65% Accurracy	270926
MNIST	image	classification	98.71% Accurracy	261206
Wine	multivariate	classification	100.0% Accurracy	444
Iris	multivariate	classification	100.0% Accurracy	150

Installation (with CMake 3.17.1)

Linux, UNIX - GCC 10.2

• To compile open a command prompt and run cmake -G"Unix Makefiles" ./.. && make from the build folder.

• To run the unit tests execute ./tests/unit_tests/UnitTests from build folder.

• To run the dataset tests run ./ImportDatasets.sh and execute ./tests/dataset_tests/DatasetTests from build folder.

Windows - MSVC++ 14.2

• You can generate a Visual Studio project by running cmake -G"Visual Studio 17 2022" ./.. from build folder.

• To run the unit tests open ./build/tests/unit_tests/UnitTests.vcxproj in Visual Studio.

• To run the dataset tests run ./build/ImportDatasets.sh and open ./build/tests/dataset_tests/DatasetTests.vcxproj in Visual Studio.

Use

Create, train and use a neural network in few lines of code.

using namespace snn;

Data data(problem::classification, inputData, expectedOutputs);

StraightforwardNeuralNetwork neuralNetwork({
    Input(1, 28, 28), // (C, X, Y)
    Convolution(16, 3, activation::ReLU), // 16 filters and (3, 3) kernels
    FullyConnected(92),
    FullyConnected(10, activation::identity, Softmax())
});

neuralNetwork.train(data, 0.90_acc || 20_s); // Train neural network on data until 90% accuracy or 20s

float accuracy = neuralNetwork.getGlobalClusteringRate(); // Retrieve the accuracy

see more details

License

Apache License 2.0