The implementations provided here do not require any toolboxes, especially no neural network toolbox.
The neural network implementations in this repo are set up in three complexities:
simpleTrainExample: Script which trains a neural network of 3 layers (in, hidden, out), each consisting of only a single neuron. Not really functional for any task, primarily available for understanding the basic behavior.trainNeuralNetStatic: Function which is capable of training a basic neural network of 3 layers with an arbitrary amount of neurons each. Look at this before looking at the next function! Takes a network generated by thegenerateNeuralNetfunction and usesapplyNeuralNet.trainNeuralNet: Function which trains networks of arbitrary size. Basically wraps the backpropagation behavior oftrainNeuralNetStaticinto an extra loop in order to handle more layers.