def test_hidden_layer_net_on_XOR(self):
network = self.setup1()
training_set = [
Instance([0.0, 0.0], [0.0]),
Instance([0.0, 1.0], [1.0]),
Instance([1.0, 0.0], [1.0]),
Instance([1.0, 1.0], [0.0])
]
# train
epoch, MSE = network.backpropagation(training_set, ERROR_LIMIT=1e-2)
assert epoch == 4098
assert MSE == 0.0099993587735512333
def test_single_layer_net_on_OR(self):
network = self.setup0()
training_set = [
Instance([0.0, 0.0], [0.0]),
Instance([0.0, 1.0], [1.0]),
Instance([1.0, 0.0], [1.0]),
Instance([1.0, 1.0], [1.0])
]
# train
epoch, MSE = network.backpropagation(training_set, ERROR_LIMIT=1e-2)
assert epoch == 340
assert MSE == 0.0099759064076452887
def test_2(self):
aTrainingExample = Instance([1.0], [0])
network = self.setup1()
network.layers[0].neurons[0].weights = [1.0, -1.0]
outputs = network.calc_networks_output(aTrainingExample.features)
result = outputs[0]
expected = 0.0
diff = abs(result - expected)
assert diff < 0.00001
def test_hidden_layer_net0(self):
aTrainingExample = Instance([2.5], [0])
network = self.setup_single_hidden_layer_net()
network.layers[0].neurons[0].weights = [1.0, 0.0]
network.layers[1].neurons[0].weights = [1.0, 0.0]
outputs = network.calc_networks_output(aTrainingExample.features)
result = outputs[0]
expected = 2.5
diff = abs(result - expected)
assert diff < 0.00001
aTrainingExample = Instance([2.5], [1.0])
network.layers[0].neurons[0].weights = [1.0, 1.0]
outputs = network.calc_networks_output(aTrainingExample.features)
result = outputs[0]
expected = 3.5
diff = abs(result - expected)
assert diff < 0.00001
def test_hidden_layer_net1(self):
aTrainingExample = Instance([2.0], [1.0])
network = self.setup_single_hidden_layer_net1()
network.layers[0].neurons[0].weights = [1.0, 1.0]
network.layers[1].neurons[0].weights = [1.0, 0.0]
network.activation_functions = [sigmoid_function, linear_function]
outputs = network.calc_networks_output(aTrainingExample.features)
result = outputs[0]
expected = 1.0 / (1.0 + math.exp(-3.0))
diff = abs(result - expected)
assert diff < 0.00001
def test_hidden_layer_net2(self):
aTrainingExample = Instance([0.5], [0.0])
network = self.setup_single_hidden_layer_net2()
network.layers[0].neurons[0].weights = [1.0, 1.0]
network.layers[1].neurons[0].weights = [1.0, 1.0]
outputs = network.calc_networks_output(aTrainingExample.features)
result = network.layers[0].neurons[0].output
expected_hidden_layer_output = 1.0 / (1.0 + math.exp(-1.5))
diff = abs(result - expected_hidden_layer_output)
assert diff < 0.00001
sum_of_weighted_inputs_to_output_neuron = expected_hidden_layer_output + 1.0
expected_output_neuron_output = 1.0 / (
1.0 + math.exp(-1.0 * sum_of_weighted_inputs_to_output_neuron))
result = outputs[0]
diff = abs(result - expected_output_neuron_output)
assert diff < 0.00001