adrianseeley/GATO.FFNN.SIGMOID.CMSE.js.html

## GATO.FFNN.SIGMOID.CMSE.js.html
<html><script>
function Neural_Network (Number_Of_Input_Neurons, Number_Of_Hidden_Layers, Number_Of_Neurons_Per_Hidden_Layer, Number_Of_Output_Neurons) {
	return [Input_Layer(Number_Of_Input_Neurons)].concat(Hidden_Layers(Number_Of_Input_Neurons, Number_Of_Hidden_Layers, Number_Of_Neurons_Per_Hidden_Layer)).concat([Output_Layer(Hidden_Layers == 0 ? Number_Of_Input_Neurons: Number_Of_Neurons_Per_Hidden_Layer, Number_Of_Output_Neurons)]);
};
function Input_Layer (Number_Of_Input_Neurons) {
	var ret = [];
	for (var n = 0; n < Number_Of_Input_Neurons; n++)
		ret.push(Input_Neuron());
	return ret;
};
function Input_Neuron () {
	return {Output: null};
};
function Hidden_Layers (Number_Of_Input_Neurons, Number_Of_Hidden_Layers, Number_Of_Neurons_Per_Hidden_Layer) {
	var ret = [];
	for (var h = 0; h < Number_Of_Hidden_Layers; h++)
		ret.push(Hidden_Layer(h == 0 ? Number_Of_Input_Neurons : Number_Of_Neurons_Per_Hidden_Layer, Number_Of_Neurons_Per_Hidden_Layer));
	return ret;
};
function Hidden_Layer (Number_Of_Neurons_In_Previous_Layer, Number_Of_Neurons_Per_Hidden_Layer) {
	var ret = [];
	for (var n = 0; n < Number_Of_Neurons_Per_Hidden_Layer; n++)
		ret.push(Hidden_Neuron(Number_Of_Neurons_In_Previous_Layer));
	return ret;
};
function Hidden_Neuron (Number_Of_Neurons_In_Previous_Layer) {
	return {Output: null, Input_Transfer_Weights: Input_Transfer_Weights(Number_Of_Neurons_In_Previous_Layer), Bias: 0};
};
function Input_Transfer_Weights (Number_Of_Neurons_In_Previous_Layer) {
	var ret = [];
	for (var n = 0; n < Number_Of_Neurons_In_Previous_Layer; n++)
		ret.push(0);
	return ret;
};
function Output_Layer (Number_Of_Neurons_In_Previous_Layer, Number_Of_Output_Neurons) {
	var ret = [];
	for (var n = 0; n < Number_Of_Output_Neurons; n++)
		ret.push(Output_Neuron(Number_Of_Neurons_In_Previous_Layer));
	return ret;
};
function Output_Neuron (Number_Of_Neurons_In_Previous_Layer) {
	return Hidden_Neuron(Number_Of_Neurons_In_Previous_Layer);
};
function Transfer (Hidden_Or_Output_Neuron, Previous_Layer) {
	Hidden_Or_Output_Neuron.Output = Hidden_Or_Output_Neuron.Bias;
	for (var n = 0; n < Previous_Layer.length; n++)
		Hidden_Or_Output_Neuron.Output += Previous_Layer[n].Output * Hidden_Or_Output_Neuron.Input_Transfer_Weights[n];
};
function Activate (Hidden_Or_Output_Neuron) {
	Hidden_Or_Output_Neuron.Output = 1 / (1 + Math.exp(-Hidden_Or_Output_Neuron.Output));
};
function Run (Neural_Network, Input_Values) {
	// Provide input values to input neurons
	for (var i = 0; i < Input_Values.length; i++)
		Neural_Network[0][i].Output = Input_Values[i];

	// Resolve network
	for (var h = 1; h < Neural_Network.length; h++) {
		for (var n = 0; n < Neural_Network[h].length; n++) {
			Transfer(Neural_Network[h][n], Neural_Network[h - 1]);
			Activate(Neural_Network[h][n]);
		}
	}

	// Gather output
	var ret = [];
	for (var n = 0; n < Neural_Network[Neural_Network.length - 1].length; n++)
		ret.push(Neural_Network[Neural_Network.length - 1][n].Output);
	return ret;
};
function Neural_Network_Fitness (Neural_Network, Cases) {
	var errors = [];
	for (var c = 0; c < Cases.length; c++) {
		var output = Run(Neural_Network, Cases[c][0]);
		var output_error = 0;
		for (var o = 0; o < output.length; o++)
			output_error += Math.pow(output[o] - Cases[c][1][o], 2);
		errors.push(output_error / output.length);
		// the output error for each case uses MSE to calculate
		// the total component error.
		// ie. for a case with 10 output components we calculate
		// the MSE of the components for that case, then add it
		// to the collected errors for all cases, which the MSE
		// is then taken of. We are effectively compounding 2 layers
		// of MSE.
	}
	var error = 0;
	for (var e = 0; e < errors.length; e++)
		error += Math.pow(errors[e], 2);
	return error / errors.length;
	// We define fitness here using MSE, which stands for
	// mean squared errors - or more simply, the average
	// of all the errors squares.
	// ie ((e1^2) + (e2^2) + (e3^2)) / 3 = mse
	// Therefore the fitness is the MSE of the collected
	// MSE's of each cases output component delta from the correct
	// output.
};

var xor = [
	[[0, 0], [0]],
	[[0, 1], [1]],
	[[1, 0], [1]],
	[[1, 1], [0]]
];
var nn = Neural_Network(2, 1, 3, 1);
Neural_Network_Fitness(nn, xor);
</script></html>
	<html><script>
	function Neural_Network (Number_Of_Input_Neurons, Number_Of_Hidden_Layers, Number_Of_Neurons_Per_Hidden_Layer, Number_Of_Output_Neurons) {
	return [Input_Layer(Number_Of_Input_Neurons)].concat(Hidden_Layers(Number_Of_Input_Neurons, Number_Of_Hidden_Layers, Number_Of_Neurons_Per_Hidden_Layer)).concat([Output_Layer(Hidden_Layers == 0 ? Number_Of_Input_Neurons: Number_Of_Neurons_Per_Hidden_Layer, Number_Of_Output_Neurons)]);
	};
	function Input_Layer (Number_Of_Input_Neurons) {
	var ret = [];
	for (var n = 0; n < Number_Of_Input_Neurons; n++)
	ret.push(Input_Neuron());
	return ret;
	};
	function Input_Neuron () {
	return {Output: null};
	};
	function Hidden_Layers (Number_Of_Input_Neurons, Number_Of_Hidden_Layers, Number_Of_Neurons_Per_Hidden_Layer) {
	var ret = [];
	for (var h = 0; h < Number_Of_Hidden_Layers; h++)
	ret.push(Hidden_Layer(h == 0 ? Number_Of_Input_Neurons : Number_Of_Neurons_Per_Hidden_Layer, Number_Of_Neurons_Per_Hidden_Layer));
	return ret;
	};
	function Hidden_Layer (Number_Of_Neurons_In_Previous_Layer, Number_Of_Neurons_Per_Hidden_Layer) {
	var ret = [];
	for (var n = 0; n < Number_Of_Neurons_Per_Hidden_Layer; n++)
	ret.push(Hidden_Neuron(Number_Of_Neurons_In_Previous_Layer));
	return ret;
	};
	function Hidden_Neuron (Number_Of_Neurons_In_Previous_Layer) {
	return {Output: null, Input_Transfer_Weights: Input_Transfer_Weights(Number_Of_Neurons_In_Previous_Layer), Bias: 0};
	};
	function Input_Transfer_Weights (Number_Of_Neurons_In_Previous_Layer) {
	var ret = [];
	for (var n = 0; n < Number_Of_Neurons_In_Previous_Layer; n++)
	ret.push(0);
	return ret;
	};
	function Output_Layer (Number_Of_Neurons_In_Previous_Layer, Number_Of_Output_Neurons) {
	var ret = [];
	for (var n = 0; n < Number_Of_Output_Neurons; n++)
	ret.push(Output_Neuron(Number_Of_Neurons_In_Previous_Layer));
	return ret;
	};
	function Output_Neuron (Number_Of_Neurons_In_Previous_Layer) {
	return Hidden_Neuron(Number_Of_Neurons_In_Previous_Layer);
	};
	function Transfer (Hidden_Or_Output_Neuron, Previous_Layer) {
	Hidden_Or_Output_Neuron.Output = Hidden_Or_Output_Neuron.Bias;
	for (var n = 0; n < Previous_Layer.length; n++)
	Hidden_Or_Output_Neuron.Output += Previous_Layer[n].Output * Hidden_Or_Output_Neuron.Input_Transfer_Weights[n];
	};
	function Activate (Hidden_Or_Output_Neuron) {
	Hidden_Or_Output_Neuron.Output = 1 / (1 + Math.exp(-Hidden_Or_Output_Neuron.Output));
	};
	function Run (Neural_Network, Input_Values) {
	// Provide input values to input neurons
	for (var i = 0; i < Input_Values.length; i++)
	Neural_Network[0][i].Output = Input_Values[i];

	// Resolve network
	for (var h = 1; h < Neural_Network.length; h++) {
	for (var n = 0; n < Neural_Network[h].length; n++) {
	Transfer(Neural_Network[h][n], Neural_Network[h - 1]);
	Activate(Neural_Network[h][n]);
	}
	}

	// Gather output
	var ret = [];
	for (var n = 0; n < Neural_Network[Neural_Network.length - 1].length; n++)
	ret.push(Neural_Network[Neural_Network.length - 1][n].Output);
	return ret;
	};
	function Neural_Network_Fitness (Neural_Network, Cases) {
	var errors = [];
	for (var c = 0; c < Cases.length; c++) {
	var output = Run(Neural_Network, Cases[c][0]);
	var output_error = 0;
	for (var o = 0; o < output.length; o++)
	output_error += Math.pow(output[o] - Cases[c][1][o], 2);
	errors.push(output_error / output.length);
	// the output error for each case uses MSE to calculate
	// the total component error.
	// ie. for a case with 10 output components we calculate
	// the MSE of the components for that case, then add it
	// to the collected errors for all cases, which the MSE
	// is then taken of. We are effectively compounding 2 layers
	// of MSE.
	}
	var error = 0;
	for (var e = 0; e < errors.length; e++)
	error += Math.pow(errors[e], 2);
	return error / errors.length;
	// We define fitness here using MSE, which stands for
	// mean squared errors - or more simply, the average
	// of all the errors squares.
	// ie ((e1^2) + (e2^2) + (e3^2)) / 3 = mse
	// Therefore the fitness is the MSE of the collected
	// MSE's of each cases output component delta from the correct
	// output.
	};

	var xor = [
	[[0, 0], [0]],
	[[0, 1], [1]],
	[[1, 0], [1]],
	[[1, 1], [0]]
	];
	var nn = Neural_Network(2, 1, 3, 1);
	Neural_Network_Fitness(nn, xor);
	</script></html>