class NNClassification(NN):
def __init__(self,
hidden_size=HIDDEN_NET_SIZE,
hfunc_type='identity',
test_function='quadratic'):
self.nn = NeuralNetwork(LEARNING_CONSTANT)
self.input_neuron_x = self.nn.add_input(None, 'x-coord')
self.input_neuron_y = self.nn.add_input(None, 'y-coord')
self.output_neuron = self.nn.add_output('above f(x)',
function='binary')
self.hidden_size = hidden_size
self.hidden_layer = []
self.hfunc = hfunc_type
self.test_function = TestFunction(test_function)
def fitness_function(self):
# TODO: should we peanalize network size
# Generate cases and build overall loss, which we would like to minimize
test_cases = []
return_cases = []
for i in range(TEST_SIZE):
x_coord = random.uniform(MIN_TEST_VALUE, MAX_TEST_VALUE)
y_coord = random.uniform(
self.test_function.evaluate(MIN_TEST_VALUE),
self.test_function.evaluate(MAX_TEST_VALUE))
if y_coord > self.test_function.evaluate(x_coord):
above = True
else:
above = False
test_cases.append([{
self.input_neuron_x: x_coord,
self.input_neuron_y: y_coord
}, above])
total_error = 0
for case in test_cases:
r_value = self.nn.feed_forward(case[0])
if r_value == 1 and case[1] != True:
total_error += 1
elif r_value == 0 and case[1] != False:
total_error += 1
return total_error / len(test_cases)
def generate_random(self):
# Randomly assign weights to the neurons
for i in range(self.hidden_size):
new_hidden = NeuralNetworkHiddenNeuron(self.hfunc)
self.input_neuron_x.add_connection(new_hidden,
random.uniform(-1, 1))
self.input_neuron_y.add_connection(new_hidden,
random.uniform(-1, 1))
self.nn.add_hidden({self.output_neuron: random.uniform(-1, 1)},
function=self.hfunc)
self.hidden_layer.append(new_hidden)
# Randomly connect some of the hidden neurons (50%)
# TODO: make sure we do not create loopy graphs, only add downstream
# connections
for i in range(self.hidden_size - 1):
# Add the ability to tweak the ratio of random connections
if random.randint(0, 1) is 1:
rand = i
while rand is i:
rand = random.randint(0, self.hidden_size - 1)
self.hidden_layer[i].add_connection(self.hidden_layer[rand],
random.uniform(-1, 1))
def test(self):
pass