def test_data(self):
'''Test data from a data file'''
print(
"*******************************TESTDATA*************************")
inputs = []
input_file = [self.bias, 7, 83, 78, 26, 71, 29.3, 0.767, 36]
for i in input_file:
inputs.append(round((float(i) / self.largest_data), 4))
d_output = 1
if self.algorithm == 0 or self.algorithm == 1:
print(
"The inputs to the network are {0} and the desired output is {1}"
.format(inputs, d_output))
temp = np.matrix(inputs)
weights = np.matrix(self.weights[0]).T
print(temp, weights)
product = np.dot(temp, weights)
if self.algorithm == 0:
activated = threshold_function(self.threshold, product.item())
else:
activated = line_equation(product.item())
print(
"The current weights and model outputs are {0} and {1} respectively"
.format(weights, activated))
else:
#backprop
#loop thru each data input
print(
"The inputs to the network are {0} and the desired output is {1}"
.format(inputs, d_output))
input_0 = np.matrix(inputs)
output_0 = sigmoid_function(input_0)
print("The outputs of the input layer are {0}".format(output_0))
weights = np.matrix(self.weights[0]).T
input_1 = np.dot(weights, input_0.T)
print("The inputs of the hidden layer are {0}".format(input_1))
output_1 = sigmoid_function(input_1)
print("The outputs of the hidden layer are {0}.".format(output_1))
input_2 = np.dot(self.weights[1], output_1)
print("The input to the output layer is {0}".format(input_2))
output_2 = sigmoid_function(input_2.item())
print("The output to the output layer is {0}".format(output_2))
def test(self):
'''Test the model to check if it conforms to
the standards'''
#true class
class_p = 1
class_n = 0
tp = 0
fn = 0
fp = 0
tn = 0
print("******************TESTING**************")
for data in self.test_set:
#loop thru each data input
inputs = data[:-1]
d_output = data[-1:][0]
print(
"The inputs to the network are {0} and the desired output is {1}"
.format(inputs, d_output))
if self.algorithm == 0 or self.algorithm == 1:
temp = np.matrix(inputs)
weights = np.matrix(self.weights[0]).T
product = np.dot(temp, weights)
if self.algorithm == 0:
m_output = threshold_function(self.threshold,
product.item())
else:
m_output = line_equation(product.item())
print("The models output is {0}".format(m_output))
else:
input_0 = np.matrix(inputs)
output_0 = line_equation(input_0)
print(
"The outputs of the input layer are {0}".format(output_0))
weights = np.matrix(self.weights[0]).T
input_1 = np.dot(weights, input_0.T)
print("The inputs of the hidden layer are {0}".format(input_1))
output_1 = sigmoid_function(input_1)
print("The outputs of the hidden layer are {0}.".format(
output_1))
input_2 = np.dot(self.weights[1], output_1)
print("The input to the output layer is {0}".format(input_2))
m_output = sigmoid_function(input_2.item())
print("The output to the output layer is {0}".format(m_output))
if m_output == d_output == class_p:
tp += 1
elif m_output == d_output == class_n:
tn += 1
elif m_output == class_p and d_output == class_n:
fp += 1
elif m_output == class_n and d_output == class_p:
fn += 1
else:
print("An error has occured")
print("**********COMPLETE***************")
print("Analysis....")
recall = tp / (tp + fn)
precision = tp / (tp + fp)
f_score = (2 * (precision * recall)) / (precision + recall)
print("The recall is: {0}".format(recall))
print("The precision is: {0}".format(precision))
print("The F-score is: {0}".format(f_score))
def backprop(self):
'''
Backprop learning and training
'''
x = 0
flag = True
#infinite loop
while flag:
#repeat the epoch until there is a convergence
print(
'******************** Start of Epoch {0} ********************'.
format(x))
errors = []
for data in self.normalized:
#loop thru each data input
inputs = data[:-1]
d_output = data[-1:][0]
print(
"The inputs to the network are {0} and the desired output is {1}"
.format(inputs, d_output))
input_0 = np.matrix(inputs)
output_0 = line_equation(input_0)
print(
"The outputs of the input layer are {0}".format(output_0))
weights = np.matrix(self.weights[0]).T
input_1 = np.dot(weights, input_0.T)
print("The inputs of the hidden layer are {0}".format(input_1))
output_1 = sigmoid_function(input_1)
print("The outputs of the hidden layer are {0}.".format(
output_1))
input_2 = np.dot(self.weights[1], output_1)
print("The input to the output layer is {0}".format(input_2))
output_2 = sigmoid_function(input_2.item())
print("The output to the output layer is {0}".format(output_2))
error = d_output - output_2
self.average_errors.append(error)
errors.append(error * error)
#calculate weight for the second weights
d = (d_output - output_2) * output_2 * (1 - output_2)
weight_change_w = self.learning_rate * d * output_1
#calculate weights for the first weights
w = np.matrix(self.weights[1]).T * d
a, b, g, h = (w[0].item(), w[1].item(), output_1[0].item(),
output_1[1].item())
weight_change_v = [[a * g * (1 - g)], [b * h * (1 - h)]]
weight_change_v = np.matrix(weight_change_v).T
#update weights
print("Updating weights.......")
self.weights[0] = ((output_0.T * weight_change_v) *
self.learning_rate) + self.weights[0]
self.weights[1] = np.matrix(
self.weights[1]).T + weight_change_w
self.weights[1] = self.weights[1].T
print("The updated weights are:")
self.output_weights()
print(
'******************** End of Epoch {0} ********************'.
format(x))
average_error = np.mean(errors)
print("The average error is, {0}".format(average_error))
if round(average_error, 1) <= self.tolerance:
print("The desired error tolerance has been achieved.")
print("Exiting... ")
flag = False
break
x += 1