init = 0.00001
d = 10
# learning rate functions
def lrf1(t):
return init / (1 + ((init / d) * t))
def lrf2(t):
return init / (1 * t)
# Question 2.2
print("SVM in primal domain with stocahstic gradient descent.")
print("\nWith first learning rate function: ")
for C in Cs:
print('\nC value = ' + str(C))
parameters = SVM.PrimalSVM(epochs, lrf1, C, train_data, train_labels)
print("Parameters:")
print(parameters)
error = SVM.test(parameters, train_data, train_labels)
print("Train error: " + str(error))
error = SVM.test(parameters, test_data, test_labels)
print("Test error: " + str(error))
input("\nWith second learning rate function: ")
for C in Cs:
print('\nC value = ' + str(C))
parameters = SVM.PrimalSVM(epochs, lrf2, C, train_data, train_labels)
print("Parameters:")
print(parameters)
error = SVM.test(parameters, train_data, train_labels)
print("Train error: " + str(error))
error = SVM.test(parameters, test_data, test_labels)
