CNet is a C++/CUDA framework for building and researching deep complex valued neural networks, as well as for optimization of complex valued functions using gradient descent with Wirtinger derivatives. In the current version it should be relatively straighforward to implement new CPU-only functions / layers. In subsequent releases will add support to easily implement CUDA layers, too.