Tezma is a small C++ machine learning framework, created for no reason. It includes the following features:

• Small tensor math library(mirroring NumPy)
• Layers(Linear, Conv2D)
• Loss(Relu, Tanh)
• Optimizer(Adam)
• Network class(adding layers, traning, prediction)
• Serialization(saving and loading of net)

Tezma was created for learning purposes.

Installation

To install

Cmake linking

Tezma supports static linking.

Usage

To use Tezma in your project, firstly clone the repo like so

$ git clone https://github.com/Dualsub/tezma.git tezma
$ cd tezma
$ mkdir build
$ cd build
$ cd cmake ..

Examples

XOR example

XOR is not a linear function and is therefore a good.

int main(int argc, char const *argv[])
{
    tz::Tensor<float> inputs = tz::tensor<float>({
        {0.0f, 0.0f}, {0.0f, 1.0f}, {1.0f, 0.0f}, {1.0f, 1.0f}
    });
    inputs.reshape({ 4, 2, 1 });
    
    tz::Tensor<float> inputs = tz::tensor<float>({
        0.0f, 1.0f, 1.0f, 0.0f
    });
    outputs.reshape({ 4, 1, 1 });
    
    // Creating model
    size_t epochs = 10000;
    
    tz::Net net;

    net.add<tz::LinearLayer>(2, 3);
    net.add<tz::Tanh>();
    net.add<tz::LinearLayer>(3, 1);
    net.add<tz::Tanh>();
    
    net.fit(inputs, outputs, MSELoss(), 0.1, epochs);

    net.saveas("nets/xor_net.tzn");

    return 0;
}

MNIST example

Mnist is a classical dataset used for ML. The implementation in Tezma is as follows:

int main(int argc, char const *argv[])
{
    tz::Dataset dataset;
    dataset.from_csv("datasets/mnist/");

    auto&[inputs, outputs] = dataset.traning_data();

    // Creating model
    size_t epochs = 10000;
    
    tz::Net<float> net;

    net.add<tz::LinearLayer>(28 * 28, 40);
    net.add<tz::Tanh>();
    net.add<tz::LinearLayer>(40, 10);
    net.add<tz::Tanh>();

    net.fit(inputs, outputs, AdamOptimizer(), 0.1, epochs);

    net.saveas("nets/mnist_net.tzn");

    return 0;
}