Feed forward backpropagation neural network implementation. Aiming to make for general use, for a network of arbitrary topology.

Currently, this implementation consists of:

• User-specificification of layer amount, and input/output/hidden layer ranks;
• Read-in of csv-type data: assumes numerical csv data, where each line contains input data followed by expected output(s) (must specify number of columns that are inputs);
• Uses a sigmoid as the transfer function.

NOTE: Forward connections exist between all possible neuron pairs.

FeedForwardNetwork object is constructed with mandatory integer array argument, which specifies the structure of the network, e.g.

FeedForwardNet network = new FeedForwardNet(new int[]{2,2,1});

creates a network with 2 input neurons, 2 hidden neurons, and a single output neuron.

Optional arguments include:

• learnRate for weight adjustments;
• lambda parameter for Sigmoid function;
• maximum iterations for learning loop;
• threshold error value for convergence.

Learn method is called with input and expected output arguments passed, each as lists of double arrays.

Still to add:

• Testing and accuracy methods;
• Generalisation of transfer functions;
• Exporting of network parameters to text file;
• Various dependency injections, generalisation and tidying up;
• Optimisation once complete;
• Probably a lot more...