Version 1.1

✔️ New optimizer: Mini batches

✔️ GPU accelerated training

✔️ Improved data structure

✔️ Different activation function for every layer

✔️ Fast saving and loading

Quick start

Example project under ./SeayonDemo

When you are using ./SeayonMnist you have to download the Mnist dataset and put it into the ./SeayonMnist/res folder.

Including in your Project

Project is built using Visual Studio Code and CMake on Windows. The code runs about 30% faster with compiled with GNU than with MSVC.

The default branch contains the currently in development beta version. The latest stable version is found under the "Release" tab. To use Seayon in your project you include the seayon.hpp header:

#include "seayon.hpp"

int main()
{
  seayon net(...);

  return 0;
}

To use the GPU accelerated version you include the cuda_seayon.cuh header and use the cuda_seayon class (use ./SeayonMnistCuda as reference). You need to install the Cuda toolkit. Follow the Cuda Installation Guide to setup the compiler.

#include "cuda_seayon.cuh"

int main()
{
  cuda_seayon net(...);

  return 0;
}

Found a bug

If you have found an issue or you would like to submit improvements, use the Issues tab. The Issues tab can also be used to start discussions. Make sure beforehand whether the issue ticket has already been created.

Contributing

I am very grateful for any contributions. Fork the repository to create your own copy, when you're done you can create a pull request.

Before opening a pull request, please submit an issue explaining the bug or feature and always reference the issue hashtag in the discription with an closing keyword: close, closes, closed, fix, fixes, fixed, resolve, resolves, resolved