def helper(partb):
dir_path = os.path.dirname(os.path.realpath(__file__))
data = BankNoteData.BankNoteData(dir_path + '/../Data/bank_note/train.csv',
shift_origin=True)
test_data = BankNoteData.BankNoteData(dir_path +
'/../Data/bank_note/test.csv',
shift_origin=True)
layer_widths = [5, 10, 25, 50, 100]
train_percentages = []
test_percentages = []
max_iters = 5
for layer_width in layer_widths:
print("Running test width hidden layer width: %i" % layer_width)
if partb:
weights = np.zeros((layer_width, layer_width, 4))
for i in range(0, layer_width):
for j in range(0, layer_width):
weights[i, j, 2] = np.random.normal()
for i in range(0, layer_width):
weights[i, 1, 3] = np.random.normal()
for i in range(0, data.features.shape[1]):
for j in range(0, layer_width):
weights[i, j, 1] = np.random.normal()
else:
weights = np.zeros((layer_width, layer_width, 4))
three_layer_nn = ThreeLayerNN.ThreeLayerNN(
num_units_per_layer=layer_width, weights=weights)
grad_descent = GradientDescent.GradientDescent(features=data.features,
labels=data.output)
results = grad_descent.fit_stochastic(
max_iters=max_iters,
args=None,
weights=weights,
step_function=training_schedule,
obj_func=three_layer_nn.objective_function,
grad_func=three_layer_nn.gradient)
train_percentage = get_percentages(data, three_layer_nn.predict)
test_percentage = get_percentages(test_data, three_layer_nn.predict)
train_percentages.append(train_percentage)
test_percentages.append(test_percentage)
print("Train error percentage was %.16f" % train_percentage)
print("Test error percentage was %.16f" % test_percentage)
t = np.linspace(0, max_iters, max_iters)
plt.plot(t, results[1])
plt.show()
plt.semilogx(layer_widths, train_percentages, label='Train Percentages')
plt.semilogx(layer_widths, test_percentages, label='Test Percentages')
plt.legend()
plt.show()
def choice_c():
dir_path = os.path.dirname(os.path.realpath(__file__))
data = BankNoteData.BankNoteData(dir_path + '/../../Data/bank_note/train.csv', shift_origin=True)
test_data = BankNoteData.BankNoteData(dir_path + '/../../Data/bank_note/test.csv', shift_origin=True)
perceptron = Perceptron.Perceptron(10, data.features, data.output, 0.5, perceptron_type='averaged')
perceptron.fit()
train_percentage = get_percentages(perceptron, data)
test_percentage = get_percentages(perceptron, test_data)
print("Weight vector was:")
print(perceptron.weights)
print("Train error percentage was %.16f" % train_percentage)
print("Test error percentage was %.16f" % test_percentage)
def helper(parta, obj_fun, grad_func):
dir_path = os.path.dirname(os.path.realpath(__file__))
data = BankNoteData.BankNoteData(
dir_path + '/../Data/bank_note/train.csv',
shift_origin=True,
)
test_data = BankNoteData.BankNoteData(dir_path +
'/../Data/bank_note/test.csv',
shift_origin=True)
if parta:
variances = np.array([0.01, 0.1, 0.5, 1, 3, 5, 10, 100])
else:
variances = [0]
max_iters = 100
train_percentages = []
test_percentages = []
for var in variances:
grad_descent = GradientDescent.GradientDescent(features=data.features,
labels=data.output)
results = grad_descent.fit_stochastic(max_iters=max_iters,
obj_func=obj_fun,
grad_func=grad_func,
step_function=training_schedule,
weights=np.zeros(
data.features.shape[1]),
args=var)
train_percentage_ = lr.get_percentages(data, results[0])
test_percentage_ = lr.get_percentages(test_data, results[0])
train_percentages.append(train_percentage_)
test_percentages.append(test_percentage_)
print("Train error percentage was %.16f" % train_percentage_)
print("Test error percentage was %.16f" % test_percentage_)
t = np.linspace(0, max_iters, max_iters)
plt.plot(t, results[1], label='Objective Function')
plt.legend()
plt.show()
if parta:
plt.semilogx(variances, train_percentages, label='Train Percentages')
plt.semilogx(variances, test_percentages, label='Test Percentages')
plt.legend()
plt.show()
def choice_b():
dir_path = os.path.dirname(os.path.realpath(__file__))
data = BankNoteData.BankNoteData(dir_path + '/../../Data/bank_note/train.csv', shift_origin=True)
test_data = BankNoteData.BankNoteData(dir_path + '/../../Data/bank_note/test.csv', shift_origin=True)
perceptron = Perceptron.Perceptron(10, data.features, data.output, 0.5, perceptron_type='voted')
perceptron.fit()
print("The vectors and their respective count are:")
np.set_printoptions(precision=4)
for i in range(0, len(perceptron.weights[0])):
print(perceptron.weights[0][i], " : %s" % (str(perceptron.weights[1][i])), )
train_percentage = get_percentages(perceptron, data)
test_percentage = get_percentages(perceptron, test_data)
print("Train error percentage was %.16f" % train_percentage)
print("Test error percentage was %.16f" % test_percentage)
def kernel_perceptron_test():
dir_path = os.path.dirname(os.path.realpath(__file__))
data = BankNoteData.BankNoteData(dir_path + '/../Data/bank_note/train.csv',
shift_origin=True)
test_data = BankNoteData.BankNoteData(dir_path +
'/../Data/bank_note/test.csv',
shift_origin=True)
gammas = [0.01, 0.1, 0.5, 1., 2., 5., 10., 100.]
for gamma in gammas:
print("Running Kernel Perceptron with Gamma = %.16f" % gamma)
kernel_perceptron = KernelPerceptron.KernelPerceptron(
features=data.features, labels=data.output, gamma=gamma)
perceptron = kernel_perceptron.fit(num_epochs=10)
train_percentage = get_percentages(data, kernel_perceptron)
test_percentage = get_percentages(test_data, kernel_perceptron)
print("Weight vector was:")
print(perceptron)
print("Train error percentage was %.16f" % train_percentage)
print("Test error percentage was %.16f" % test_percentage)
def choice_d():
dir_path = os.path.dirname(os.path.realpath(__file__))
data = BankNoteData.BankNoteData(dir_path + '/../../Data/bank_note/train.csv', shift_origin=True)
test_data = BankNoteData.BankNoteData(dir_path + '/../../Data/bank_note/test.csv', shift_origin=True)
train_errors = [[], [], []]
test_errors = [[], [], []]
print("Calculating statistics")
sys.stdout.write('Progress: [%s]' % (' ' * 100))
sys.stdout.flush()
for i in range(1, 101):
standard_perceptron = Perceptron.Perceptron(10, data.features, data.output, 0.5)
standard_perceptron.fit()
voted_perceptron = Perceptron.Perceptron(10, data.features, data.output, 0.5, perceptron_type='voted')
voted_perceptron.fit()
averaged_perceptron = Perceptron.Perceptron(10, data.features, data.output, 0.5, perceptron_type='averaged')
averaged_perceptron.fit()
train_errors[0].append(get_percentages(standard_perceptron, data))
train_errors[1].append(get_percentages(voted_perceptron, data))
train_errors[2].append(get_percentages(averaged_perceptron, data))
test_errors[0].append(get_percentages(standard_perceptron, test_data))
test_errors[1].append(get_percentages(voted_perceptron, test_data))
test_errors[2].append(get_percentages(averaged_perceptron, test_data))
sys.stdout.write('\rProgress: [%s' % ('#' * i))
sys.stdout.write('%s]' % (' ' * (100 - i)))
sys.stdout.flush()
print("\n%24s %10s %10s" % ("Standard", "Voted", "Averaged"))
print("Train Errors: %.8f %.8f %.8f" % (float(np.mean(np.array(train_errors[0]))),
float(np.mean(np.array(train_errors[1]))),
float(np.mean(np.array(train_errors[2])))))
print("Test Errors: %.8f %.8f %.8f" % (float(np.mean(np.array(test_errors[0]))),
float(np.mean(np.array(test_errors[1]))),
float(np.mean(np.array(test_errors[2])))))