Joining Networks

[sdesalas]

Some thoughts:

• Currently this replicates the behaviour of Array.prototype.concat() however a better way is that network1.append(network2) means that network1 absorbs network2, so its not possible to use it. or if its used, all calls refer to the portion of the new network (since the neurons are attached by ref). This makes more sense as in most cases networks will be created on the spot (ie network.append(new NeuralNetwork(100)).
• Several options (signalSpeed, learningRate, learningPeriod) make more sense at the neuron level, rather than for the whole network (though be careful of performance impact of too many objects during serialization). Reason for this is that animal brains have sections that are specialised in short term memory, and other sections specialise in long term memory, and not every neuron propagates signals at the same speed. Other options however (connectionsPerNeuron, shape) only make sense during network instantiation and should not be stored at the network level.
• Ideally, it should be easy to 'build' your network via the interface, adding new networks and changing the existing network in-place. This is where joining networks would become really useful.