APPENDIX C – Test Parameters

1. Neural Network

1) Test ww1-network-2:

2) Test ww1-network-3:

backpropagation step constant = 0.1

number of epochs = 1000

2. Modular Network

1) Test ww1-modular-4:

expert backpropagation step constant = 0.1

training algorithm = winner-takes-all

step constant = 0.1

memory constant = 0.1

improvement constant = 1.0

number of epochs = 1000

3. Genetic Algorithm

1) Test ww1-breeder-1:

size of population = 6

max. number of layers = 3

max. number of nodes per layer = 5

fitness evaluator = LinearEvaluator

selection algorithm = FitnessProportionate

crossover operation = CrossoverOperation1

mutation algorithm = MutateAll

number of epochs of training before evaluation = 200

fitness threshold = 0.95

percentage replaced = 50

percentage mutated = 10

2) Test ww1-breeder-4:

size of population = 6

max. number of layers = 3

max. number of nodes per layer = 5

fitness evalutaor = SquareEvaluator

selection algorithm = FitnessProportionate

crossover operation = CrossoverOperation1

mutation algorithm = MutateNodes

number of epochs of training before evaluation = 100

fitness threshold = 0.9

percentage replaced = 50

percentage mutated = 50

4. Modular Breeder

number of modular networks = 5

desired error = 0.1

percentage improvement that breeder must achieve = 30

module training step constant = 0.1

modular network training memory constant = 0.1

modular network training step constant = 0.1

modular network training improvement constant = 1.0

percentage of population to replace = 50

1) Test ww1-modbreeder-3:

initial number of modules in each network = 3

maximum number of layers in randomly generated networks = 3

maximum number of nodes per layer in randomly generated networks = 6

number of epochs to train modular networks = 100

number of epochs to train each population member = 100

percentage of population to mutate = 20

2) Test ww1-modbreeder-4

initial number of modules in each network = 3

maximum number of layers in randomly generated networks = 3

maximum number of nodes per layer in randomly generated networks = 6

number of epochs to train modular networks = 100

number of epochs to train each population member = 100

percentage of population to mutate = 20

3) Test ww1-modbreeder-6

initial number of modules in each network = 3

maximum number of layers in randomly generated networks = 2

maximum number of nodes per layer in randomly generated networks = 5

number of epochs to train modular networks = 50

number of epochs to train each population member = 50

percentage of population to mutate = 50