This is a self contained k-layer neural network implementation with gradient descent and backpropagation, inspired by a previous 2 layer NN implementation: https://github.com/sertaco/Two-Layer-Neural-Network-NN2.git
% NNk implements a k-layer neural network with variable input, hidden and % output layer sizes. NNk trains this NN for the given sample matrix % X and output matrix y. NNk can be used for multiclass classification. % The number of layers is fixed to three. NNk utilizes backpropagation with % gradient descent.
%Inputs: %*X: training set with nXm dimensions, m: no of training samples,n: no of features %*y: training output with rXm dimensions, r: no of classes %hidden_layer_size: [a b c ...] e.g., b being the layer size of the second %hidden layer. The number of elements of the size array determines the %number of hidden layers (k) %*MaxIter: maximum iteration in gradient descent of each training %*lambda: regularization parameter.
%Needed m files: % -randInitializeWeights.m % -nnkCostFunction.m % -sigmoidGradient.m % -fmincg.m % -houtk.m % -sigmoid.m % -gettheta.m % -predictNNk.m