def test_should_add_itself_to_other_10x10_matrix(self):
matrix = Matrix(10, 10, 1)
other_matrix = Matrix(10, 10, 1336)
matrix.add(other_matrix)
self.assertEquals(1337, matrix.get_item(9, 9))
class OutputLayer(Layer):
def __init__(self, size):
Layer.__init__(self, size)
self.previous_layer = None
self.W = None
self.B = None
self.E = None
self.activation_function = lambda x: 1 / (1 + math.exp(-x))
def initialize(self, previous_layer):
self.previous_layer = previous_layer
self.W = Matrix(self.size, previous_layer.size).randomize(-1, 1)
self.B = Matrix(self.size, 1).randomize(-1, 1)
def feed_forward(self):
self.values = ((self.W * self.previous_layer.values) +
self.B).map_function(self.activation_function)
def calculate_errors(self, target_arr):
if len(target_arr) == self.size:
self.E = Matrix.from_list(target_arr, self.size, 1) - self.values
else:
raise ValueError("Incorrect target size.")
def adjust_parameters(self, learning_rate):
gradients = self.values.map_function(
lambda x: x * (1 - x)).get_hadamard_product(
self.E).get_scalar_multiple(learning_rate)
self.W.add(gradients * self.previous_layer.values.get_transpose())
self.B.add(gradients)
def test_should_add_itself_to_other_1x1_matrix(self):
matrix = Matrix(1, 1, 1)
other_matrix = Matrix(1, 1, 1336)
matrix.add(other_matrix)
self.assertEquals(1337, matrix.get_item(0, 0))
def test_add_scalar_to_matrix(self):
m = Matrix(3, 3)
m.data[0] = [1, 2, 3]
m.data[1] = [4, 5, 6]
m.data[2] = [7, 8, 9]
m.add(1)
self.assertEqual(m.rows, 3)
self.assertEqual(m.cols, 3)
self.assertEqual(m.data, [[2, 3, 4], [5, 6, 7], [8, 9, 10]])
def feed_forward(self, input_list):
inputs = Matrix.fromList(input_list)
print(inputs)
hidden = Matrix.multiply(self.weights_ih, inputs)
Matrix.add(hidden, self.bias_h)
hidden.map_matrix(sigmoid)
output = Matrix.multiply(self.weights_ho, hidden)
Matrix.add(output, self.bias_o)
output.map_matrix(sigmoid)
return output
def test_matrix_add():
from matrix import Matrix
m_1 = Matrix((4, 3))
m_1[0][1] = 2
m_2 = Matrix((2, 2), value=[[1, 2], [3, 4]])
with pytest.raises(Exception) as _:
_ = Matrix.add(m_1, m_2)
m_3 = Matrix((4, 3), value=[[1, 2, 3], [0, 2, 1], [4, 5, -1], [1, 2, 0]])
m_add = Matrix.add(m_1, m_3)
assert m_add[0][0] == 1
assert m_add[0][1] == 4
assert m_add[0][2] == 3
def test_add_matrix_to_matrix(self):
m = Matrix(2, 2)
m.data[0] = [1, 2]
m.data[1] = [3, 4]
n = Matrix(2, 2)
n.data[0] = [10, 11]
n.data[1] = [12, 13]
m.add(n)
self.assertEqual(m.rows, 2)
self.assertEqual(m.cols, 2)
self.assertEqual(m.data, [[11, 13], [15, 17]])
def parse(string):
if string in name_space:
value = name_space.get(string)
print value
else:
mat = Matrix()
l = string.replace(' ', '')
if l.find('=') != -1:
if l[2:] not in name_space and l.find('zeros') != -1:
zer = mat.matrix_creation_using_zeros(l[2:])
name_space[str(l[0])] = zer
elif l[3] not in name_space:
matr = mat.matrix_creation(l[2:])
name_space[str(l[0])] = matr
elif l[3] and l[5] in name_space:
if (l.find(',') != -1) or (l.find(';') != -1):
conc = mat.concat(l[2:])
name_space[str(l[0])] = conc
if l.find('+') != -1:
add = mat.add(l)
print add
if l.find("\'") != -1:
tr = mat.transposer(l)
print tr
if l.find('inv') != -1:
inv = mat.inverser(l[-2])
print inv
def test_add_matrix():
passed = True
failed_tests = []
options = [[[1, 2, 3], [2, 3, 4], [3, 4, 5]],
[[0, 0, 0], [0, 0, 0], [0, 0, 0]]]
for o_index, option in enumerate(options):
test_matrix_a = Matrix.build_from_rows([[1, 2, 3], [1, 2, 3],
[1, 2, 3]])
test_matrix_b = Matrix.build_from_rows(option)
matrices_to_add = [test_matrix_a, test_matrix_b]
correct_values = []
for i in range(test_matrix_a.dimensions[0]):
correct_values.append([])
for j in range(test_matrix_a.dimensions[1]):
correct_values[i].append(0)
for k in range(len(matrices_to_add)):
correct_values[i][j] += matrices_to_add[k].values[i][j]
result_matrix = Matrix.add(matrices_to_add)
for i in range(result_matrix.dimensions[0]):
for j in range(result_matrix.dimensions[1]):
if result_matrix.values[i][j] != correct_values[i][j]:
passed = False
failed_tests.append(o_index)
if not passed:
print(
'TEST FAILED: Add matrices failed tests: {}.'.format(failed_tests))
else:
print('TEST PASSED: Add matrices test passed.')
def test_add_matrix_initial():
passed = True
failed_tests = []
options = [
[[1, 2], [1, 2]],
[[1, 2, 3], [1, 2, 3]],
]
for o_index, option in enumerate(options):
test_matrix_a = Matrix.build_from_rows([[1, 2, 3], [1, 2, 3]])
test_matrix_b = Matrix.build_from_rows(option)
result_matrix = Matrix.add(
[test_matrix_a, test_matrix_a, test_matrix_a, test_matrix_b])
if o_index != len(options) - 1:
if result_matrix.valid:
passed = False
failed_tests.append(o_index)
else:
if not result_matrix.valid:
passed = False
failed_tests.append(o_index)
if not passed:
print('TEST FAILED: Add matrices initial failed tests: {}.'.format(
failed_tests))
else:
print('TEST PASSED: Add matrices initial test passed.')
def test_add_None(self):
mat_a = Matrix(2,2)
mat_b = Matrix(3,2)
mat_a.map(lambda x: x+1)
mat_b.map(lambda x: x+2)
mat_ret_none = mat_a.add(mat_b)
self.assertNotIsInstance(type(mat_ret_none), Matrix)
def test_add(self):
a = [[1,2,3],[3,5,6],[6,1,1]]
b = [[4,2,5],[1,7,5],[3,9,2]]
A = Matrix([3,3], elems = a)
B = Matrix([3,3], elems = b)
correct = np.add(a,b)
self.assertEqual(A.add(B).elems[1][1] , correct[1][1])
def test_add_scalar(self):
mat = Matrix(2,2)
mat.map(lambda x: x+1)
mat_ret = mat.add(2)
self.assertEqual(mat_ret.rows, 2)
self.assertEqual(mat_ret.cols, 2)
self.assertEqual(mat_ret.data, [[3,3], [3,3]])
def main():
A = Matrix([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
# B = Matrix([1,2,3])
# C = Matrix([1, [2, 3]])
B = Matrix([[10, 2, 3], [4, 50, 6]])
print(A)
print(B)
print(A.add(B))
def test(): matrix_1 = Matrix(4,5,6,7) matrix_2 = Matrix(2,2,2,1) matrix_3 = matrix_2.add(matrix_1) matrix_4 = matrix_3.prod(matrix_2) print(matrix_3) print(matrix_4) matrix_1[0][1] = 23 print(matrix_1)
def test_add_elementwise(self):
mat_a = Matrix(2,2)
mat_b = Matrix(2,2)
mat_a.map(lambda x: x+1)
mat_b.map(lambda x: x+2)
mat_ret = mat_a.add(mat_b)
self.assertEqual(mat_ret.rows, 2)
self.assertEqual(mat_ret.cols, 2)
self.assertEqual(mat_ret.data, [[3,3], [3,3]])
class HiddenLayer(Layer):
def __init__(self, size):
Layer.__init__(self, size)
self.previous_layer = None
self.next_layer = None
self.W = None
self.B = None
self.E = None
self.activation_function = lambda x: 1 / (1 + math.exp(-x))
def initialize(self, previous_layer, next_layer):
self.previous_layer = previous_layer
self.next_layer = next_layer
self.W = Matrix(self.size, previous_layer.size).randomize(-1, 1)
self.B = Matrix(self.size, 1).randomize(-1, 1)
def feed_forward(self):
self.values = (self.W * self.previous_layer.values +
self.B).map_function(self.activation_function)
def calculate_errors(self):
self.E = self.next_layer.W.get_transpose() * self.next_layer.E
def adjust_parameters(self, learning_rate):
# print(self.values)
# print(self.E)
gradients = self.values.map_function(
lambda x: x * (1 - x)).get_hadamard_product(
self.E).get_scalar_multiple(learning_rate)
# print(gradients)
self.W.add(gradients * self.previous_layer.values.get_transpose())
self.B.add(gradients)
def add(self, a, b):
if isinstance(a, float):
if isinstance(b, float):
return a + b
elif isinstance(b, Complex):
return Complex.add(Complex(a), b)
elif isinstance(b, Matrix):
raise ComputorException('Illegal operation: Rational + Matrix')
elif isinstance(a, Complex):
if isinstance(b, float):
return Complex.add(a, Complex(b))
elif isinstance(b, Complex):
return Complex.add(a, b)
elif isinstance(b, Matrix):
raise ComputorException('Illegal operation: Complex + Matrix')
elif isinstance(a, Matrix):
if isinstance(b, float):
raise ComputorException('Illegal operation: Matrix + Rational')
elif isinstance(b, Complex):
raise ComputorException('Illegal operation: Matrix + Complex')
elif isinstance(b, Matrix):
return Matrix.add(a, b)
raise ComputorException('Computor.add(): something bad happened 🤷')
class NeuralNetwork:
def __init__(self, inputs, hidden, output):
self.num_inputs = inputs
self.num_hidden = hidden
self.num_output = output
# Weights randomize
self.weights_ih = Matrix(self.num_hidden, self.num_inputs)
self.weights_ho = Matrix(self.num_output, self.num_hidden)
self.weights_ih.randomize()
self.weights_ho.randomize()
# Bias randomize
self.bias_h = Matrix(self.num_hidden, 1)
self.bias_o = Matrix(self.num_output, 1)
self.bias_h.randomize()
self.bias_o.randomize()
self.learning_rate = 0.1
@staticmethod
def sigmoid(x):
return (1 / (1 + math.exp(-x)))
@staticmethod
def dsigmoid(y):
return (y * (1 - y))
def feedforward(self, input_array):
# Transform input array into inputs matrix
inputs = Matrix.static_fromArray(input_array)
# Generating the hidden layer
hidden = Matrix.static_multiply(self.weights_ih, inputs)
hidden.add(self.bias_h)
hidden.map(self.sigmoid)
# Generating the output layer
output = Matrix.static_multiply(self.weights_ho, hidden)
output.add(self.bias_o)
output.map(self.sigmoid)
# Return output array
return Matrix.static_toArray(output)
def train(self, input_array, target_array):
# Transform input array into inputs matrix
inputs = Matrix.static_fromArray(input_array)
self.last_input = inputs
# Generating the hidden layer
hidden = Matrix.static_multiply(self.weights_ih, inputs)
hidden.add(self.bias_h)
hidden.map(self.sigmoid)
# Generating the output layer
output = Matrix.static_multiply(self.weights_ho, hidden)
output.add(self.bias_o)
output.map(self.sigmoid)
# Transform target array into target Matrix
target = Matrix.static_fromArray(target_array)
# 1. Calculate the errors
## a. Output errors
output_errors = Matrix.static_substract(target, output)
# 2. Calculate the gradient
## a. Output gradient (gradient of our output cost function)
output_gradient = Matrix.static_map(output, self.dsigmoid)
output_gradient.multiply(output_errors)
# 3. Calculate the deltas adjustment
## a. Output deltas
hidden_t = Matrix.static_transpose(hidden)
weights_ho_delta = Matrix.static_multiply(output_gradient, hidden_t)
weights_ho_delta.multiply(self.learning_rate)
# 4. Adjust the original weights
## a. Output weights & output bias
weights_ho = self.weights_ho
self.weights_ho.add(weights_ho_delta)
self.bias_o.add(output_gradient)
# 1. Calculate the errors
## b. Hidden layer errors
who_t = Matrix.static_transpose(self.weights_ho)
hidden_errors = Matrix.static_multiply(who_t, output_errors)
# 2. Calculate the gradient
## b. Hidden gradient (gradient of our hidden cost function)
hidden_gradient = Matrix.static_map(hidden, self.dsigmoid)
hidden_gradient = Matrix.static_map(hidden, self.dsigmoid)
hidden_gradient.multiply(hidden_errors)
# 3. Calculate the deltas adjustment
## b. Hidden deltas
inputs_t = Matrix.static_transpose(inputs)
weights_ih_delta = Matrix.static_multiply(hidden_gradient, inputs_t)
weights_ih_delta.multiply(self.learning_rate)
self.weights_ih_delta = weights_ih_delta
# 4. Adjust the original weights
## b. Biases weights & hidden bias
self.weights_ih.add(weights_ih_delta)
self.bias_h.add(hidden_gradient)
class NeuralNetwork:
num_of_input_nurons = None
num_of_hiddden_nurons = None
num_of_output_nurons = None
error = 0
lr = 0.1
#initialize the neural network layers, weights and biases
def __init__(self,
no_of_input,
no_of_hidden,
no_of_output,
weight_file="weights.txt"):
user_cancel = False
if (weight_file == ""):
user_cancel = True
self.num_of_input_nurons = no_of_input
self.num_of_hiddden_nurons = no_of_hidden
self.num_of_output_nurons = no_of_output
#check if exist the weights of the local drive
isWeightsExist = Path(weight_file).exists()
if (isWeightsExist and not user_cancel):
f = open(weight_file, "r")
f1 = f.readlines()
w = []
w_count = 0
b = []
b_count = 0
w_i_h = []
w_h_o = []
b_h = []
b_o = []
for text in f1:
if (text.find("w")) != -1:
w.append(ast.literal_eval(text.split(":")[1]))
elif (text.find("b")) != -1:
b.append(ast.literal_eval(text.split(":")[1]))
for i in range(no_of_hidden):
w_i_h.append([])
for j in range(no_of_input):
w_i_h[i].append(w[w_count])
w_count = w_count + 1
for i in range(no_of_hidden):
b_h.append([])
for j in range(1):
b_h[i] = b[b_count]
b_count = b_count + 1
for i in range(no_of_output):
w_h_o.append([])
for j in range(no_of_hidden):
w_h_o[i].append(w[w_count])
w_count = w_count + 1
for i in range(no_of_output):
b_o.append([])
for j in range(1):
b_o[i] = b[b_count]
b_count = b_count + 1
self.weight_iput_to_hidden = Matrix.array_to_matrix(
w_i_h, no_of_hidden, no_of_input)
self.weight_hidden_to_output = Matrix.array_to_matrix(
w_h_o, no_of_output, no_of_hidden)
self.bias_of_hidden = Matrix.array_to_vector(b_h)
self.bias_of_output = Matrix.array_to_vector(b_o)
else:
self.weight_iput_to_hidden = Matrix(no_of_hidden, no_of_input)
self.weight_hidden_to_output = Matrix(no_of_output, no_of_hidden)
self.weight_iput_to_hidden.randomize()
self.weight_hidden_to_output.randomize()
self.bias_of_hidden = Matrix(no_of_hidden, 1)
self.bias_of_output = Matrix(no_of_output, 1)
self.bias_of_hidden.randomize()
self.bias_of_output.randomize()
#sigmoid activation function
@staticmethod
def sigmoid(x):
return 1 / (1 + math.exp(-x))
#derivatives of the sigmoid activation function
@staticmethod
def d_sigmoid(y):
return y * (1 - y)
#this is the method which predict the result
def predict(self, input_array):
inputs = Matrix.array_to_vector(input_array)
hidden = Matrix.dot_product(self.weight_iput_to_hidden, inputs)
hidden.add(self.bias_of_hidden)
hidden.map(NeuralNetwork.sigmoid)
output = Matrix.dot_product(self.weight_hidden_to_output, hidden)
output.add(self.bias_of_output)
output.map(NeuralNetwork.sigmoid)
return output.to_array()
#this is the method which train the network with backpropagation in gradient decent algorithm
def train(self, input_array, target_array):
#converts the input array to a vector
inputs = Matrix.array_to_vector(input_array)
hidden = Matrix.dot_product(self.weight_iput_to_hidden, inputs)
hidden.add(self.bias_of_hidden)
hidden.map(NeuralNetwork.sigmoid)
output = Matrix.dot_product(self.weight_hidden_to_output, hidden)
output.add(self.bias_of_output)
output.map(NeuralNetwork.sigmoid)
target_Matrix = Matrix.array_to_vector(target_array)
#calculate error
error = Matrix.sub(target_Matrix, output)
self.error = error
weight_hidden_to_output_t = Matrix.transpose(
self.weight_hidden_to_output)
hidden_error = Matrix.dot_product(weight_hidden_to_output_t, error)
#calculate the gradient of the hidden layer
hidden_gradient_matrix = Matrix.map_s(output, NeuralNetwork.d_sigmoid)
hidden_gradient_matrix.scale(error)
hidden_gradient_matrix.scale(self.lr)
delta_weights_ho = Matrix.dot_product(output, Matrix.transpose(hidden))
#update weights and biases
self.weight_hidden_to_output.add(delta_weights_ho)
self.bias_of_output.add(hidden_gradient_matrix)
#calculate the gradient of the output layer
output_gradient_matrix = Matrix.map_s(hidden, NeuralNetwork.d_sigmoid)
output_gradient_matrix.scale(hidden_error)
output_gradient_matrix.scale(self.lr)
delta_weights_ih = Matrix.dot_product(hidden, Matrix.transpose(inputs))
#update weights and biases
self.bias_of_hidden.add(output_gradient_matrix)
self.weight_iput_to_hidden.add(delta_weights_ih)
class NeuralNetwork:
def __init__(self, input_nodes, hidden_nodes=False, output_nodes=False):
if not hidden_nodes:
a = input_nodes
self.input_nodes = a.input_nodes
self.hidden_nodes = a.hidden_nodes
self.output_nodes = a.output_nodes
self.weights_ih = a.weights_ih.copy()
self.weights_ho = a.weights_ho.copy()
self.weights_ho_t = a.weights_ho_t.copy()
self.bias_h = a.bias_h.copy()
self.bias_o = a.bias_o.copy()
else:
self.input_nodes = input_nodes
self.hidden_nodes = hidden_nodes
self.output_nodes = output_nodes
self.weights_ih = Matrix(self.hidden_nodes, self.input_nodes)
self.weights_ho = Matrix(self.output_nodes, self.hidden_nodes)
self.weights_ho_t = Matrix.transpose(self.weights_ho)
self.bias_h = Matrix(self.hidden_nodes, 1)
self.bias_o = Matrix(self.output_nodes, 1)
self.weights_ih.randomize()
self.weights_ho.randomize()
self.bias_h.randomize()
self.bias_o.randomize()
self.learning_rate = 0.1
def predict(self, input_array):
# Computing Hidden Outputs
inputs = Matrix.fromArray(input_array)
hidden = Matrix.multiply1(self.weights_ih, inputs)
hidden.add(self.bias_h)
hidden.map1(sigmoid) # Activation Function
# Computing Output Layer's Output!
outputs = Matrix.multiply1(self.weights_ho, hidden)
outputs.add(self.bias_o)
outputs.map1(sigmoid)
return outputs.toArray()
def train(self, input_array, target_array):
# Computing Hidden Outputs
inputs = Matrix.fromArray(input_array)
hidden = Matrix.multiply1(self.weights_ih, inputs)
hidden.add(self.bias_h)
hidden.map1(sigmoid) # Activation function
# Computing Output Layer's Output!
outputs = Matrix.multiply1(self.weights_ho, hidden)
outputs.add(self.bias_o)
outputs.map1(sigmoid) # Neural Net's Guess
# Converting target array to matrix object
targets = Matrix.fromArray(target_array)
# Calculate Error
output_errors = Matrix.subtract(targets, outputs)
# Calculate Hidden Errors
hidden_errors = Matrix.multiply1(self.weights_ho_t, output_errors)
# Calculate gradients
gradients = Matrix.map2(outputs, dsigmoid)
gradients.multiply2(output_errors)
gradients.multiply2(self.learning_rate)
# Calculate Deltas
hidden_t = Matrix.transpose(hidden)
weight_ho_deltas = Matrix.multiply1(gradients, hidden_t)
# Adjust Hidden -> Output weights and output layer's biases
self.weights_ho.add(weight_ho_deltas)
self.bias_o.add(gradients)
# Calculate the hidden gradients
hidden_gradients = Matrix.map2(hidden, dsigmoid)
hidden_gradients.multiply2(hidden_errors)
hidden_gradients.multiply2(self.learning_rate)
# Calculate Deltas
input_t = Matrix.transpose(inputs)
weight_ih_deltas = Matrix.multiply1(hidden_gradients, input_t)
# Adjust Input -> Hidden weights and hidden biases
self.weights_ih.add(weight_ih_deltas)
self.bias_h.add(hidden_gradients)
def mutate(self, rate):
def mutate(x):
return x + randomGaussian(0, 0.1) if random() < rate else x
self.weights_ih.map1(mutate)
self.weights_ho.map1(mutate)
self.bias_h.map1(mutate)
self.bias_o.map1(mutate)
def copy(self):
return NeuralNetwork(self)
class NeuralNetwork:
def __init__(self, input_nodes, hidden_nodes, output_nodes):
self.input_nodes = input_nodes
self.hidden_nodes = hidden_nodes
self.output_nodes = output_nodes
self.learning_rate = 0.1
self.weights_ih = Matrix(self.hidden_nodes, self.input_nodes)
self.weights_ho = Matrix(self.output_nodes, self.hidden_nodes)
self.weights_ih.randomize()
self.weights_ho.randomize()
self.bias_h = Matrix(self.hidden_nodes, 1)
self.bias_o = Matrix(self.output_nodes, 1)
self.bias_h.randomize()
self.bias_o.randomize()
def predict(self, input_array):
# Gerando as saídas intermediárias
inputs = Matrix.fromArray(input_array)
hidden = Matrix.multiply(self.weights_ih, inputs)
hidden.add(self.bias_h)
# Função de ativação
hidden.map(sigmoid)
# Gerando as saídas de facto
output = Matrix.multiply(self.weights_ho, hidden)
output.add(self.bias_o)
output.map(sigmoid)
return output.toArray()
def feedfoward(self, input_array):
# GERANDO OS HIDDEN OUTPUTS (SAIDAS DOS NÓS INTERMEDIÁRIOS)
# Transforma a lista de entrada em um objeto vetor (ou matriz unidimensional, como quiser)
input = Matrix.fromArray(input_array)
# Multiplica as entradas pelos respectivos pesos
hidden = Matrix.multiply(self.weights_ih, input)
# Adiciona o bias
hidden.add(self.bias_h)
# Função de ativação!
hidden.map(sigmoid)
# GERANDO A SAÍDA
# Multiplica as saidas dos nós intermediários pelos respectivos pesos
output = Matrix.multiply(self.weights_ho, hidden)
output.add(self.bias_o)
output.map(sigmoid)
# Transforma o objeto vetor em uma lista
return output.toArray() # E manda de volta! hehe ;)
def train(self, input_array, target_array): # Backpropagation
# GERANDO OS HIDDEN OUTPUTS (SAIDAS DOS NÓS INTERMEDIÁRIOS)
# Transforma a lista de entrada em um objeto vetor (ou matriz unidimensional, como quiser)
inputs = Matrix.fromArray(input_array)
# Multiplica as entradas pelos respectivos pesos
hidden = Matrix.multiply(self.weights_ih, inputs)
# Adiciona o bias
hidden.add(self.bias_h)
# Função de ativação!
hidden.map(sigmoid)
# GERANDO A SAÍDA
# Multiplica as saidas dos nós intermediários pelos respectivos pesos
outputs = Matrix.multiply(self.weights_ho, hidden)
outputs.add(self.bias_o)
outputs.map(sigmoid)
# Coverte para uma matriz objeto
targets = Matrix.fromArray(target_array)
# Calcula o erro
# ERRO = ALVOS - SAIDAS
output_errors = Matrix.subtract(targets, outputs)
# Calcula o gradiente
gradients = Matrix.mapIt(outputs, dsigmoid)
gradients.multiplyBy(output_errors)
gradients.multiplyBy(self.learning_rate)
# Calcula deltas
hidden_T = Matrix.transpose(hidden)
weight_ho_deltas = Matrix.multiply(gradients, hidden_T)
# Ajusta os pesos pelos deltas
self.weights_ho.add(weight_ho_deltas)
# Ajusta os bias pelos seus deltas
self.bias_o.add(gradients)
# Calcula os erros da camada intermediária
who_t = Matrix.transpose(self.weights_ho)
hidden_errors = Matrix.multiply(who_t, output_errors)
# Calcula o gradiente intermediário
hidden_gradient = Matrix.mapIt(hidden, dsigmoid)
hidden_gradient.multiplyBy(hidden_errors)
hidden_gradient.multiplyBy(self.learning_rate)
# Calcula o delta intermediário
inputs_T = Matrix.transpose(inputs)
weight_ih_deltas = Matrix.multiply(hidden_gradient, inputs_T)
self.weights_ih.add(weight_ih_deltas)
# Ajusta os bias pelos seus deltas
self.bias_h.add(hidden_gradient)
def copy(self):
return self
def mutate(self, rate):
def mutateIt(val):
if random.random() < rate:
return val + random.gauss(0, 0.1)
else:
return val
self.weights_ih.map(mutateIt)
self.weights_ho.map(mutateIt)
self.bias_h.map(mutateIt)
self.bias_o.map(mutateIt)
#Good luck.
from matrix import Matrix
if __name__ == "__main__":
matrix_square_ascending = Matrix(1, 2, 3, 4)
matrix_square = Matrix(2, 2, 2, 2)
matrix_cubic_ascending = Matrix(1, 2, 3, 4, 5, 6, 7, 8, 9)
matrix_cubic = Matrix(1, 1, 1, 1, 1, 1, 1, 1, 1)
print("Add:")
print(matrix_cubic_ascending)
print(matrix_cubic)
print("Result:")
matrix_added = matrix_cubic_ascending.add(matrix_cubic)
print(matrix_added)
print("------------------------\n")
print("Subtract:")
print(matrix_square_ascending)
print(matrix_square)
print("Result:")
matrix_subtracted = matrix_square_ascending.subtract(matrix_square)
print(matrix_subtracted)
print("------------------------\n")
print("Dummy multiply:")
print(matrix_square_ascending)
print(matrix_square)
print("Result:")
class NeuralNetwork():
def __init__(self, number_inputs, number_hiddens, number_outputs):
self.number_inputs = number_inputs
self.number_hiddens = number_hiddens
self.number_outputs = number_outputs
self.weights_ih = Matrix(rows=self.number_hiddens,
cols=self.number_inputs)
self.weights_ho = Matrix(rows=self.number_outputs,
cols=self.number_hiddens)
self.bias_h = Matrix(rows=self.number_hiddens, cols=1)
self.bias_h.map(lambda x: 1.0)
self.bias_o = Matrix(rows=self.number_outputs, cols=1)
self.bias_o.map(lambda x: 1.0)
self.learning_rate = 0.1
# initialize the Weights with random values
self.weights_ih.randomize()
self.weights_ho.randomize()
def feedforward(self, input):
# Extracts the output array from the 3-tuple returned by self.__guess
return self.__guess(input)[0]
def __guess(self, input):
in_matrix = Matrix.fromList(input)
hidden = Matrix.product(self.weights_ih, in_matrix)
hidden.add(self.bias_h)
hidden.map(self.__activate)
output = Matrix.product(self.weights_ho, hidden)
output.add(self.bias_o)
output.map(self.__activate)
return (output.toList(), hidden, in_matrix)
def train(self, inputs, target_label):
guess_r = self.__guess(
inputs) # (output as list, hidden, input matrix)
guess = Matrix.fromList(guess_r[0])
hidden = guess_r[1]
input_matrix = guess_r[2]
target_matrix = Matrix.fromList(target_label)
# Calculate output errors
output_errors = Matrix.subtract(target_matrix, guess)
# Calculating gradients for Hidden -> output
gradients_ho = Matrix(guess.data)
gradients_ho.map(self.__activate_derivative)
gradients_ho.multiply(output_errors)
gradients_ho.multiply(self.learning_rate)
# Calculating deltas
weights_ho_deltas = Matrix.product(gradients_ho,
Matrix.transpose(hidden))
# Tweaking weights_ho with the calculated deltas
self.weights_ho.add(weights_ho_deltas)
# Tweaking hidden -> output bias with the gradients
self.bias_o.add(gradients_ho)
# Calculate hidden layer errors
hidden_errors = Matrix.product(Matrix.transpose(self.weights_ho),
output_errors)
# Calculating gradients for Input -> Hidden
gradients_ih = Matrix(hidden.data)
gradients_ih.map(self.__activate_derivative)
gradients_ih.multiply(hidden_errors)
gradients_ih.multiply(self.learning_rate)
# Calculating deltas
weights_ih_deltas = Matrix.product(gradients_ih,
Matrix.transpose(input_matrix))
# Tweaking weights_ih with the calculated deltas
self.weights_ih.add(weights_ih_deltas)
# Twaeking input -> hidden bias with the gradients
self.bias_h.add(gradients_ih)
def __activate(self, val):
# Activate uses Sigmoid function
# https://en.wikipedia.org/wiki/Sigmoid_function
return 1.0 / (1 + math.exp(-val))
def __activate_derivative(self, active_val):
return active_val * (1 - active_val)
class NeuralNetwork:
def __init__(self, numI, numH, numO):
self.input_nodes = numI
self.hidden_nodes = numH
self.output_nodes = numO
self.weights_ih = Matrix(self.hidden_nodes, self.input_nodes)
self.weights_ho = Matrix(self.output_nodes, self.hidden_nodes)
self.weights_ih.randomize()
self.weights_ho.randomize()
self.bias_h = Matrix(self.hidden_nodes, 1)
self.bias_o = Matrix(self.output_nodes, 1)
self.bias_h.randomize()
self.bias_o.randomize()
self.learning_rate = 0.3
def feedforward(self, inputs):
# Generate hidden output
hiddens = MatMath.multiply(self.weights_ih, inputs)
hiddens.add(self.bias_h)
# Activation function on hidden
hiddens.map(sigmoid)
# Generate output
outputs = MatMath.multiply(self.weights_ho, hiddens)
outputs.add(self.bias_o)
# Activation function on output
outputs.map(sigmoid)
return (outputs, hiddens)
def guess(self, input_arr):
inputs = Matrix.fromArray(input_arr)
(outputs, _) = self.feedforward(inputs)
return outputs.toArray()
def train(self, input_arr, target_arr):
inputs = Matrix.fromArray(input_arr)
(outputs, hiddens) = self.feedforward(inputs)
targets = Matrix.fromArray(target_arr)
# Calculate the output error
# ERROR = TARGETS - OUTPUTS
output_errors = MatMath.subtract(targets, outputs)
# Calculate output gradients
output_gradients = MatMath.map(outputs, dsigmoid)
output_gradients.multiply(output_errors)
output_gradients.multiply(self.learning_rate)
# Calculate deltas
hiddens_T = MatMath.transpose(hiddens)
weights_ho_deltas = MatMath.multiply(output_gradients, hiddens_T)
# Adding deltas
self.weights_ho.add(weights_ho_deltas)
self.bias_o.add(output_gradients)
# Calculate the hidden layer errors
weights_ho_T = MatMath.transpose(self.weights_ho)
hidden_errors = MatMath.multiply(weights_ho_T, output_errors)
# Calculate hidden gradients
hidden_gradients = MatMath.map(hiddens, dsigmoid)
hidden_gradients.multiply(hidden_errors)
hidden_gradients.multiply(self.learning_rate)
# Calculate deltas
inputs_T = MatMath.transpose(inputs)
weights_ih_deltas = MatMath.multiply(hidden_gradients, inputs_T)
# Adding deltas
self.weights_ih.add(weights_ih_deltas)
self.bias_h.add(hidden_gradients)
def main(self):
# parse the command-line arguments
args = self.parser().parse_args()
file_name = args.arff
learner_name = args.L
eval_method = args.E[0]
eval_parameter = args.E[1] if len(args.E) > 1 else None
print_confusion_matrix = args.verbose
normalize = args.normalize
random.seed(args.seed) # Use a seed for deterministic results, if provided (makes debugging easier)
# load the model
learner = self.get_learner(learner_name)
# load the ARFF file
data = Matrix()
data.load_arff(file_name)
if normalize:
print("Using normalized data")
data.normalize()
# print some stats
print("\nDataset name: {}\n"
"Number of instances: {}\n"
"Number of attributes: {}\n"
"Learning algorithm: {}\n"
"Evaluation method: {}\n".format(file_name, data.rows, data.cols, learner_name, eval_method))
if eval_method == "training":
print("Calculating accuracy on training set...")
features = Matrix(data, 0, 0, data.rows, data.cols-1)
labels = Matrix(data, 0, data.cols-1, data.rows, 1)
confusion = Matrix()
start_time = time.time()
learner.train(features, labels)
elapsed_time = time.time() - start_time
print("Time to train (in seconds): {}".format(elapsed_time))
accuracy = learner.measure_accuracy(features, labels, confusion)
print("Training set accuracy: " + str(accuracy))
if print_confusion_matrix:
print("\nConfusion matrix: (Row=target value, Col=predicted value)")
confusion.print()
print("")
elif eval_method == "static":
print("Calculating accuracy on separate test set...")
test_data = Matrix(arff=eval_parameter)
if normalize:
test_data.normalize()
print("Test set name: {}".format(eval_parameter))
print("Number of test instances: {}".format(test_data.rows))
features = Matrix(data, 0, 0, data.rows, data.cols-1)
labels = Matrix(data, 0, data.cols-1, data.rows, 1)
start_time = time.time()
learner.train(features, labels)
elapsed_time = time.time() - start_time
print("Time to train (in seconds): {}".format(elapsed_time))
train_accuracy = learner.measure_accuracy(features, labels)
print("Training set accuracy: {}".format(train_accuracy))
test_features = Matrix(test_data, 0, 0, test_data.rows, test_data.cols-1)
test_labels = Matrix(test_data, 0, test_data.cols-1, test_data.rows, 1)
confusion = Matrix()
test_accuracy = learner.measure_accuracy(test_features, test_labels, confusion)
print("Test set accuracy: {}".format(test_accuracy))
if print_confusion_matrix:
print("\nConfusion matrix: (Row=target value, Col=predicted value)")
confusion.print()
print("")
elif eval_method == "random":
print("Calculating accuracy on a random hold-out set...")
train_percent = float(eval_parameter)
if train_percent < 0 or train_percent > 1:
raise Exception("Percentage for random evaluation must be between 0 and 1")
print("Percentage used for training: {}".format(train_percent))
print("Percentage used for testing: {}".format(1 - train_percent))
data.shuffle()
train_size = int(train_percent * data.rows)
train_features = Matrix(data, 0, 0, train_size, data.cols-1)
train_labels = Matrix(data, 0, data.cols-1, train_size, 1)
test_features = Matrix(data, train_size, 0, data.rows - train_size, data.cols-1)
test_labels = Matrix(data, train_size, data.cols-1, data.rows - train_size, 1)
start_time = time.time()
learner.train(train_features, train_labels)
elapsed_time = time.time() - start_time
print("Time to train (in seconds): {}".format(elapsed_time))
train_accuracy = learner.measure_accuracy(train_features, train_labels)
print("Training set accuracy: {}".format(train_accuracy))
confusion = Matrix()
test_accuracy = learner.measure_accuracy(test_features, test_labels, confusion)
print("Test set accuracy: {}".format(test_accuracy))
if print_confusion_matrix:
print("\nConfusion matrix: (Row=target value, Col=predicted value)")
confusion.print()
print("")
elif eval_method == "cross":
print("Calculating accuracy using cross-validation...")
folds = int(eval_parameter)
if folds <= 0:
raise Exception("Number of folds must be greater than 0")
print("Number of folds: {}".format(folds))
reps = 1
sum_accuracy = 0.0
elapsed_time = 0.0
for j in range(reps):
data.shuffle()
for i in range(folds):
begin = int(i * data.rows / folds)
end = int((i + 1) * data.rows / folds)
train_features = Matrix(data, 0, 0, begin, data.cols-1)
train_labels = Matrix(data, 0, data.cols-1, begin, 1)
test_features = Matrix(data, begin, 0, end - begin, data.cols-1)
test_labels = Matrix(data, begin, data.cols-1, end - begin, 1)
train_features.add(data, end, 0, data.cols - 1)
train_labels.add(data, end, data.cols - 1, 1)
start_time = time.time()
learner.train(train_features, train_labels)
elapsed_time += time.time() - start_time
accuracy = learner.measure_accuracy(test_features, test_labels)
sum_accuracy += accuracy
print("Rep={}, Fold={}, Accuracy={}".format(j, i, accuracy))
elapsed_time /= (reps * folds)
print("Average time to train (in seconds): {}".format(elapsed_time))
print("Mean accuracy={}".format(sum_accuracy / (reps * folds)))
else:
raise Exception("Unrecognized evaluation method '{}'".format(eval_method))
from matrix import Matrix matrix_1 = Matrix(4, 5, 6, 7) matrix_2 = Matrix(2, 2, 2, 1) matrix_3 = matrix_2.add(matrix_1) matrix_4 = matrix_2.multiply(matrix_1) matrix_3.print() matrix_4.print()
class neural_network():
def __init__(self, input_nodes: int, hidden_nodes: int, output_nodes: int):
self.input_nodes = input_nodes
self.hidden_nodes = hidden_nodes
self.output_nodes = output_nodes
self.weights_ih = Matrix(self.hidden_nodes, self.input_nodes)
self.weights_ho = Matrix(self.output_nodes, self.hidden_nodes)
self.weights_ih.randomize()
self.weights_ho.randomize()
self.bias_h = Matrix(self.hidden_nodes, 1)
self.bias_o = Matrix(self.output_nodes, 1)
self.bias_h.randomize()
self.bias_o.randomize()
self.set_learning_rate()
self.set_activation_function()
def copy(self):
nn = neural_network(self.input_nodes, self.hidden_nodes,
self.output_nodes)
nn.weights_ih = self.weights_ih.copy()
nn.weights_ho = self.weights_ho.copy()
nn.bias_h = self.bias_h.copy()
nn.bias_o = self.bias_o.copy()
nn.set_learning_rate()
nn.set_activation_function()
return nn
def predict(self, input_list: list) -> list:
# Generating the Hidden Outputs
inputs = Matrix.from_list(input_list)
hidden = Matrix.static_multiply(self.weights_ih, inputs)
hidden.add(self.bias_h)
# Activation function!
hidden.map(self.activation_function.x)
# Generating the output's output!
output = Matrix.static_multiply(self.weights_ho, hidden)
output.add(self.bias_o)
output.map(self.activation_function.x)
# Sending back to the caller!
return output.to_list()
def set_learning_rate(self, learning_rate: float = 0.1):
self.learning_rate = learning_rate
def set_activation_function(
self, func: activation_function = activation_function.sigmoid()):
self.activation_function = func
##own
def mutate(self, rate: float):
def func(val, i, j):
if random.uniform(0, 1) < rate:
return val + random.uniform(-0.07, 0.07)
else:
return val
self.weights_ih.map(func)
self.weights_ho.map(func)
self.bias_h.map(func)
self.bias_o.map(func)
return
def train(self, input_list: list, target_list: list):
# Generating the Hidden Outputs
inputs = Matrix.from_list(input_list)
hidden = Matrix.static_multiply(self.weights_ih, inputs)
hidden.add(self.bias_h)
# activation function!
hidden.map(self.activation_function.x)
# Generating the output's output!
outputs = Matrix.static_multiply(self.weights_ho, hidden)
outputs.add(self.bias_o)
outputs.map(self.activation_function.x)
# Convert array to matrix object
targets = Matrix.from_list(target_list)
# Calculate the error
# ERROR = TARGETS - OUTPUTS
output_errors = Matrix.subtract(targets, outputs)
# Calculate gradient
gradients = Matrix.static_map(outputs, self.activation_function.y)
gradients.multiply(output_errors)
gradients.multiply(self.learning_rate)
# Calculate deltas
hidden_T = Matrix.transpose(hidden)
weight_ho_deltas = Matrix.static_multiply(gradients, hidden_T)
# Calculate the hidden layer errors
who_t = Matrix.transpose(self.weights_ho)
hidden_errors = Matrix.static_multiply(who_t, output_errors)
# Calculate hidden gradient
hidden_gradient = Matrix.static_map(hidden, self.activation_function.y)
hidden_gradient.multiply(hidden_errors)
hidden_gradient.multiply(self.learning_rate)
# Calcuate input->hidden deltas
inputs_T = Matrix.transpose(inputs)
weight_ih_deltas = Matrix.static_multiply(hidden_gradient, inputs_T)
self.weights_ih.add(weight_ih_deltas)
# Adjust the bias by its deltas(which is just the gradients)
self.bias_h.add(hidden_gradient)
# Adjust the weights by deltas
self.weights_ho.add(weight_ho_deltas)
# Adjust the bias by its deltas(which is just the gradients)
self.bias_o.add(gradients)
def serialize(self) -> bytes:
return dill.dumps(self)
@staticmethod
def deserialize(data: bytes) -> 'neural_network':
return dill.loads(data)
class NeuralNetwork():
def __init__(self, input_nodes, hidden_nodes, output_nodes):
self.input_nodes = input_nodes
self.hidden_nodes = hidden_nodes
self.output_nodes = output_nodes
# set the learning rate
self.lr = 0.1
# inputs -> hidden layer -> outputs
self.weights_ih = Matrix(self.hidden_nodes, self.input_nodes)
self.weights_ho = Matrix(self.output_nodes, self.hidden_nodes)
# biases for hidden and output layer
self.bias_h = Matrix(self.hidden_nodes, 1)
self.bias_o = Matrix(self.output_nodes, 1)
# randomize weigts and biases
self.weights_ih.randomize()
self.weights_ho.randomize()
self.bias_h.randomize()
self.bias_o.randomize()
''' sets the learning rate '''
def setLR(lr):
self.lr = lr
def confusion_matrix(self, xs, ys, labels, normalize=False):
''' generate a confusion matrix '''
''' cols are the predictions and the rows are the vals '''
cm = Matrix(len(labels), len(labels))
for x, y in zip(xs, ys):
index_prediction, index_actual = self.get_index_val(x, y)
# print(f'{index_actual}, {index_prediction}')
cm.data[index_actual][index_prediction] += 1
if normalize:
cm.multiply(1/len(xs))
return cm
def get_index_val(self, x, y):
prediction = self.feed_forward(x)
index_prediction = prediction.index(max(prediction))
index_actual = y.index(max(y))
return index_prediction, index_actual
def compare_index_val(self, x, y):
''' compare indicies of individual points '''
index_prediction, index_actual = self.get_index_val(x, y)
if (index_prediction == index_actual):
return True
else:
return False
''' computes the accuracy '''
def get_accuracy(self, xs, ys):
''' compare the values of the indicies of the arrays '''
score = 0
for x, y in zip(xs, ys):
if self.compare_index_val(x, y):
score += 1
return score / len(xs)
''' predict '''
def predict(self, input_array):
return self.feed_forward(input_array)
''' predicts output '''
def feed_forward(self, input_array):
# get the values for the hidden nodes
inputs = Matrix.fromArray(input_array)
hidden = Matrix.matMultiply(self.weights_ih, inputs)
hidden.add(self.bias_h)
# activation function on the hidden nodes
hidden.map(sigmoid)
# get the values for the output
output = Matrix.matMultiply(self.weights_ho, hidden) # multiply ho wegths with hidden nodes
output.add(self.bias_o) # add bias
output.map(sigmoid) # sigmoid activation
# TODO: apply softmax layer
# return output as array
return output.toArray()
''' trains the nn '''
def train(self, input_array, target_array):
inputs = Matrix.fromArray(input_array)
targets = Matrix.fromArray(target_array)
# get values for hidden nodes
hidden = Matrix.matMultiply(self.weights_ih, inputs)
hidden.add(self.bias_h)
hidden.map(sigmoid)
# get values for output nodes
outputs = Matrix.matMultiply(self.weights_ho, hidden)
outputs.add(self.bias_o)
outputs.map(sigmoid)
# TODO: apply softmax layer
# calculate the error, ERROR = TARGETS - OUTPUTS
output_errors = Matrix.subtract(targets, outputs)
# get derivative of output_errors: gradien = ouptuts * (1 - outputs)
gradients = Matrix.static_map(outputs, dsigmoid)
gradients.multiply(output_errors)
gradients.multiply(self.lr)
# calculate deltas: delta_w = error * hidden_t * lr
hidden_T = Matrix.transpose(hidden)
weights_ho_deltas = Matrix.matMultiply(gradients, hidden_T)
# adjsut the weights
self.weights_ho.add(weights_ho_deltas)
# adjust bias
self.bias_o.add(gradients)
# hidden layer errors
# weights hidden out tranpose
who_t = Matrix.transpose(self.weights_ho)
hidden_errors = Matrix.matMultiply(who_t, output_errors)
# hidden gradient
hidden_gradient = Matrix.static_map(hidden, dsigmoid)
hidden_gradient.multiply(hidden_errors)
hidden_gradient.multiply(self.lr)
# calc input -> hidden deltas
inputs_T = Matrix.transpose(inputs)
weights_ih_deltas = Matrix.matMultiply(hidden_gradient, inputs_T)
# adjust weights for input_hidden
self.weights_ih.add(weights_ih_deltas)
self.bias_h.add(hidden_gradient)
from matrix import Matrix from vector import Vector matrix = Matrix(4) matrix.add(Vector.fromList([7, -5, 6, -7])) matrix.add(Vector.fromList([2, -3, 10, 9])) matrix.add(Vector.fromList([-5, 4, -2, 2])) matrix.add(Vector.fromList([8, -9, 7, 15])) m = matrix.simplify() # m.printMe() matrix = Matrix(4) matrix.add(Vector.fromList([12, -9, -6, 4])) matrix.add(Vector.fromList([-7, 4, 11, -6])) matrix.add(Vector.fromList([11, -8, -7, 10])) matrix.add(Vector.fromList([-9, 7, 3, -5])) matrix.add(Vector.fromList([5, -3, -9, 12])) m = matrix.simplify() # m.printMe() matrix = Matrix(3) matrix.add(Vector.fromList([0, 3, 3])) matrix.add(Vector.fromList([3, -7, -9])) matrix.add(Vector.fromList([-6, 8, 12])) matrix.add(Vector.fromList([6, -5, -9])) matrix.add(Vector.fromList([4, 8, 6])) matrix.add(Vector.fromList([-5, 9, 15])) m = matrix.simplify() # m.printMe() # Matrix.create([[1,3],[3,9],[4,7],[7,6]]).simplify().printMe()
class NeuralNetwork():
def __init__(self, input_nodes, hidden_nodes, output_nodes):
self.input_nodes = input_nodes
self.hidden_nodes = hidden_nodes
self.output_nodes = output_nodes
# set the learning rate
self.lr = 0.1
# inputs -> hidden layer -> outputs
self.weights_ih = Matrix(self.hidden_nodes, self.input_nodes)
self.weights_ho = Matrix(self.output_nodes, self.hidden_nodes)
# biases for hidden and output layer
self.bias_h = Matrix(self.hidden_nodes, 1)
self.bias_o = Matrix(self.output_nodes, 1)
# randomize weigts and biases
self.weights_ih.randomize()
self.weights_ho.randomize()
self.bias_h.randomize()
self.bias_o.randomize()
''' sets the learning rate '''
def setLR(lr):
self.lr = lr
''' predicts output '''
def feed_forward(self, input_array):
# get the values for the hidden nodes
inputs = Matrix.fromArray(input_array)
hidden = Matrix.matMultiply(self.weights_ih, inputs)
hidden.add(self.bias_h)
# activation function on the hidden nodes
hidden.map(sigmoid)
# get the values for the output
output = Matrix.matMultiply(self.weights_ho, hidden) # multiply ho wegths with hidden nodes
output.add(self.bias_o) # add bias
output.map(sigmoid) # sigmoid activation
# return output as array
return output.toArray()
''' trains the nn '''
def train(self, input_array, target_array):
inputs = Matrix.fromArray(input_array)
targets = Matrix.fromArray(target_array)
# get values for hidden nodes
hidden = Matrix.matMultiply(self.weights_ih, inputs)
hidden.add(self.bias_h)
hidden.map(sigmoid)
# get values for output nodes
outputs = Matrix.matMultiply(self.weights_ho, hidden)
outputs.add(self.bias_o)
outputs.map(sigmoid)
# calculate the error, ERROR = TARGETS - OUTPUTS
output_errors = Matrix.subtract(targets, outputs)
# get derivative of output_errors: gradien = ouptuts * (1 - outputs)
gradients = Matrix.static_map(outputs, dsigmoid)
gradients.multiply(output_errors)
gradients.multiply(self.lr)
# calculate deltas: delta_w = error * hidden_t * lr
hidden_T = Matrix.transpose(hidden)
weights_ho_deltas = Matrix.matMultiply(gradients, hidden_T)
# adjsut the weights
self.weights_ho.add(weights_ho_deltas)
# adjust bias
self.bias_o.add(gradients)
# hidden layer errors
# weights hidden out tranpose
who_t = Matrix.transpose(self.weights_ho)
hidden_errors = Matrix.matMultiply(who_t, output_errors)
# hidden gradient
hidden_gradient = Matrix.static_map(hidden, dsigmoid)
hidden_gradient.multiply(hidden_errors)
hidden_gradient.multiply(self.lr)
# calc input -> hidden deltas
inputs_T = Matrix.transpose(inputs)
weights_ih_deltas = Matrix.matMultiply(hidden_gradient, inputs_T)
# adjust weights for input_hidden
self.weights_ih.add(weights_ih_deltas)
self.bias_h.add(hidden_gradient)
class NeuralNetwork:
def __init__(self, input_nodes, hidden_nodes, output_nodes):
self.input_nodes = input_nodes
self.hidden_nodes = hidden_nodes
self.output_nodes = output_nodes
self.weights_ih = Matrix(self.hidden_nodes, self.input_nodes)
self.weights_ho = Matrix(self.output_nodes, self.hidden_nodes)
self.weights_ih.randomize()
self.weights_ho.randomize()
self.bias_h = Matrix(self.hidden_nodes, 1)
self.bias_o = Matrix(self.output_nodes, 1)
self.bias_h.randomize()
self.bias_o.randomize()
self.learning_rate = 0.1
# @staticmethod
# def sigmoid(*args):
# for x in args:
# return 1/(1+Math.exp(-x))
def feed_forward(self, input_array):
#Generating the hidden outputs
inputs = Matrix.fromArray(input_array)
hidden = Matrix.multiply(self.weights_ih, inputs)
hidden.add(self.bias_h)
#activation function
# print("--------------------")
# print("----------------")
hidden.mapper(sigmoid)
output = Matrix.multiply(self.weights_ho, hidden)
output.add(self.bias_o)
output.mapper(sigmoid)
return output.toArray()
def train(self, input_array, target_array):
"""------FEED_FORWARD------"""
inputs = Matrix.fromArray(input_array)
hidden = Matrix.multiply(self.weights_ih, inputs)
hidden.add(self.bias_h)
#activation function
# print("--------------------")
# print("----------------")
hidden.mapper(sigmoid)
outputs = Matrix.multiply(self.weights_ho, hidden)
outputs.add(self.bias_o)
outputs.mapper(sigmoid)
"""-----TRAINING-------"""
#Convert array to Matrix object
# outputs = Matrix.fromArray(outputs)
targets = Matrix.fromArray(target_array)
output_errors = Matrix.subtract(targets, outputs)
#Calculate output gradient
gradients = Matrix.map(outputs, dsigmoid)
gradients.scalar_multiply(output_errors)
gradients.scalar_multiply(self.learning_rate)
hiddden_T = Matrix.transpose(hidden)
weights_ho_deltas = Matrix.multiply(gradients, hiddden_T)
#Adjust the weight delatas
self.weights_ho.add(weights_ho_deltas)
#Adjust the bias
self.bias_o.add(gradients)
who_t = Matrix.transpose(self.weights_ho)
#Calculate the hidden layer errors
hidden_errors = Matrix.multiply(who_t, output_errors)
#Calculate hidden gradient
hidden_gradient = Matrix.map(hidden, dsigmoid)
hidden_gradient.scalar_multiply(hidden_errors)
hidden_gradient.scalar_multiply(self.learning_rate)
inputs_T = Matrix.transpose(inputs)
weights_ih_deltas = Matrix.multiply(hidden_gradient, inputs_T)
self.weights_ih.add(weights_ih_deltas)
self.bias_h.add(hidden_gradient)
class NeuralNetwork:
def __init__(self, in_nodes, hid_nodes, out_nodes):
if type(in_nodes) == NeuralNetwork:
a = in_nodes
self.input_nodes = a.input_nodes
self.hidden_nodes = a.hidden_nodes
self.output_nodes = a.output_nodes
self.weights_ih = a.weights_ih.copy()
self.weights_ho = a.weights_ho.copy()
self.bias_h = a.bias_h.copy()
self.biah_o = a.bias_o.copy()
else:
self.input_nodes = in_nodes
self.hidden_nodes = hid_nodes
self.output_nodes = out_nodes
self.weights_ih = Matrix(self.hidden_nodes, self.input_nodes)
self.weights_ho = Matrix(self.output_nodes, self.hidden_nodes)
self.weights_ih.randomize()
self.weights_ho.randomize()
self.bias_h = Matrix(self.hidden_nodes, 1)
self.bias_o = Matrix(self.output_nodes, 1)
self.bias_h.randomize()
self.bias_o.randomize()
self.set_learning_rate()
self.set_activation_function()
def predict(self, input_array):
inputs = Matrix.from_array(input_array)
hidden = Matrix.static_multiply(self.weights_ih, inputs)
hidden.add(self.bias_h)
hidden.map(self.activation_function.func)
output = Matrix.static_multiply(self.weights_ho, hidden)
output.add(self.bias_o)
output.map(self.activation_function.func)
return output.to_array()
def set_learning_rate(self, learning_rate=0.1):
self.learning_rate = learning_rate
def set_activation_function(self, func=sigmoid):
self.activation_function = func
def train(self, input_array, target_array):
inputs = Matrix.from_array(input_array)
hidden = Matrix.static_multiply(self.weights_ih, inputs)
hidden.add(self.bias_h)
hidden.map(self.activation_function.func)
outputs = Matrix.static_multiply(self.weights_ho, hidden)
outputs.add(self.bias_o)
outputs.map(self.activation_function.func)
targets = Matrix.from_array(target_array)
output_errors = Matrix.subtract(targets, outputs)
gradients = Matrix.static_map(outputs,
self.activation_function.dfunc)
gradients.multiply(output_errors)
gradients.multiply(self.learning_rate)
hidden_T = Matrix.transpose(hidden)
weight_ho_deltas = Matrix.static_multiply(gradients, hidden_T)
self.weights_ho.add(weight_ho_deltas)
self.bias_o.add(gradients)
who_T = Matrix.transpose(self.weights_ho)
hidden_errors = Matrix.static_multiply(who_T, output_errors)
hidden_gradient = Matrix.static_map(hidden,
self.activation_function.dfunc)
hidden_gradient.multiply(hidden_errors)
hidden_gradient.multiply(self.learning_rate)
inputs_T = Matrix.transpose(inputs)
weight_ih_deltas = Matrix.static_multiply(hidden_gradient,
inputs_T)
self.weights_ih.add(weight_ih_deltas)
self.bias_h.add(hidden_gradient)
def copy(self):
return NeuralNetwork(self)
def mutate(func):
self.weights_ih.map(func)
self.weights_ho.map(func)
self.bias_h.map(func)
self.biah_o.map(func)
