This is a header-only library.
The code in this header file is put inside a namespace datetime which contain classes Date, Time and DateTime.
The code here is in namespace Delay and features two functions sleep(T us) and update_loop(int fps, std::function<bool(void)> update_func).
This part of the library will probably move out to its own repo one day.
The standard stuff work fairly well AFAIK, but I do have some problems with the LeNet-5 implementation (see below).
The reason I wanted to implement LeNet-5 (by Yann LeCun) is to learn convolutional networks. My opinion is that you don't learn how these more advanced types of network work just by using PyTorch or TensorFlow etc, but you rather learn how to use those APIs. This is my attempt to actually understand how LeNet-5 works under the hood so that I can implement more advanced networks such as AlexNet. I've planned to add other types of machine learning techniques and networks such as LSTMs and Q-learning (RL) in the future.
The LeNet-5 code is a W.I.P. at the moment. The paper by Yann LeCun is a bit unclear on the details. The feed forward part is fairly straight forward, but issues start popping up when attempting to backprop the thing.