anima_neural_network

#Anima

##A JS neural network

In progress. Backwards compatibility not guaranteed.

Anima lets you train feedforward neural networks, with or without convolutional layers. It consists of a neuron class, which has most of the functionnality, and a network class, which provides useful functions for iterating over layers consisting of instances of the neuron class. The below documentation is suggestive rather than exaustive, tentative, and very possibly temporary.

Neuron

var neuronMaker = require('./neuron');

var inputNeuron = neuron({ typeOfNeuron: 'input' });

var neuronInstance = neuron({ typeOfNeuron: 'leakyrelu', randomness: 'flatProportionateZero' });

var secondNeuron = neuron({ typeOfNeuron: 'leakyrelu', cost: 'squaredError'

})

//Hook them up and initialize them

neuronInstance.connect(inputNeuron); secondNeuron.connect(neuronInstance); inputNeuron.init(); neuronInstance.init(); secondNeuron.init();

//Set input for neuron

inputNeuron.activation(1); neuronInstance.activate(); secondNeuron.activate();

//Set what the neuron should be and propogate error back

secondNeuron.propogate(1) neuronInstance.propogate();

//Adjust

secondNeuron.adjust(0.1); neuronInstance.adjust(0.1);