Writing a deep neural network from scratch and with keras for multiclass classification on MNIST
For my own education I tried implementing a fully connected neural network / MLP from scratch with numpy for performing a multiclass classifier task on the MNIST dataset. Writing a neural network from scratch forces to really engage with the topic in order to solve and understand it instead of depending on high level APIs such as Tensorflow.
I also implemented an MLP with keras to compare the effort/amount of LoC for basically the same task.