def test_two(in_sample, out_sample):
target_function = random_target_function()
training_set = random_set(in_sample, target_function)
weight = linear_percepton(training_set.z, training_set.y)
in_error = weight_error(weight, training_set.z, training_set.y)
testing_set = random_set(out_sample, target_function)
out_error = weight_error(weight, testing_set.z, testing_set.y)
return in_error, out_error
def test_two(in_sample, out_sample):
target_function = random_target_function()
training_set = random_set(in_sample, target_function)
weight = linear_percepton(training_set.z, training_set.y)
in_error = weight_error(weight, training_set.z, training_set.y)
testing_set = random_set(out_sample, target_function)
out_error = weight_error(weight, testing_set.z, testing_set.y)
return in_error, out_error
def test_four(in_sample, out_sample):
training_set = random_set(in_sample, moved_circle)
noisy_indices = np.random.choice(in_sample, size=round(0.1 * in_sample), replace=False)
training_set.y[noisy_indices] *= -1
weight = linear_percepton(training_set.z, training_set.y)
in_error_no_transform = weight_error(weight, training_set.z, training_set.y)
training_set.z = second_order(training_set.z)
weight = linear_percepton(training_set.z, training_set.y)
in_error_transform = weight_error(weight, training_set.z, training_set.y)
testing_set = random_set(out_sample, moved_circle, second_order)
noisy_indices = np.random.choice(out_sample, size=round(0.1 * out_sample), replace=False)
testing_set.y[noisy_indices] *= -1
out_error_transform = weight_error(weight, testing_set.z, testing_set.y)
return in_error_no_transform, weight, out_error_transform
def test1(training_data, testing_data):
training_set = DataML(training_data, transform)
weight = linear_percepton(training_set.z, training_set.y)
testing_set = DataML(testing_data, transform)
in_error, out_error = [ weight_error(weight, data_set.z, data_set.y)
for data_set in [training_set, testing_set] ]
return in_error, out_error
def trial(in_sample, out_sample):
target_function = random_target_function()
training_set = random_set(in_sample, target_function)
initial_weight = np.zeros(len(training_set.x[0]))
weight, iterations = pla(training_set.z, training_set.y, initial_weight, True)
testing_set = random_set(out_sample, target_function)
out_error = weight_error(weight, testing_set.z, testing_set.y)
return out_error, iterations
def best_model(model_weights, testing_set):
errors = [
weight_error(weight, testing_set.z[:, :len(weight)], testing_set.y)
for weight in model_weights
]
return len(
model_weights[np.argmin(errors)]
) - 1, errors # return k value that yields least error. see k_values
def trial(in_sample, out_sample):
target_function = random_target_function()
training_set = random_set(in_sample, target_function)
initial_weight = np.zeros(len(training_set.x[0]))
weight, iterations = pla(training_set.z, training_set.y, initial_weight,
True)
testing_set = random_set(out_sample, target_function)
out_error = weight_error(weight, testing_set.z, testing_set.y)
return out_error, iterations
def trial(training_data, testing_data, a):
training_set = DataML(training_data, transform)
weights = minimize_error_aug(training_set.z, training_set.y, a)
in_error, out_error = [
weight_error(weights, tset.z, tset.y)
for tset in [training_set,
DataML(testing_data, transform)]
]
return in_error, out_error
def test1(training_data, testing_data):
training_set = DataML(training_data, transform)
weight = linear_percepton(training_set.z, training_set.y)
testing_set = DataML(testing_data, transform)
in_error, out_error = [
weight_error(weight, data_set.z, data_set.y)
for data_set in [training_set, testing_set]
]
return in_error, out_error
def test_four(in_sample, out_sample):
training_set = random_set(in_sample, moved_circle)
noisy_indices = np.random.choice(in_sample,
size=round(0.1 * in_sample),
replace=False)
training_set.y[noisy_indices] *= -1
weight = linear_percepton(training_set.z, training_set.y)
in_error_no_transform = weight_error(weight, training_set.z,
training_set.y)
training_set.z = second_order(training_set.z)
weight = linear_percepton(training_set.z, training_set.y)
in_error_transform = weight_error(weight, training_set.z, training_set.y)
testing_set = random_set(out_sample, moved_circle, second_order)
noisy_indices = np.random.choice(out_sample,
size=round(0.1 * out_sample),
replace=False)
testing_set.y[noisy_indices] *= -1
out_error_transform = weight_error(weight, testing_set.z, testing_set.y)
return in_error_no_transform, weight, out_error_transform
def myTrial(in_sample, out_sample):
target_function = random_target_function()
#w0 es agregado por default
training_set = random_set(in_sample, target_function)
initial_weight = np.zeros(len(training_set.x[0]))
weight, iterations = myOwnPlaImplementation(training_set.z, training_set.y,
initial_weight)
testing_set = random_set(out_sample, target_function)
out_error = weight_error(weight, testing_set.z, testing_set.y)
return out_error, iterations
def trial(in_sample, out_of_sample):
target_function = random_target_function()
training_set = random_set(in_sample, target_function)
pla_weight = pla(training_set.z, training_set.y)
svm_weight = svm(training_set.z, training_set.y)
testing_set = random_set(out_of_sample, target_function)
pla_error, svm_error = [ weight_error(weight, testing_set.z, testing_set.y)
for weight in
[ pla_weight, svm_weight] ]
def helper(x):
if x > 0:
return 0
else:
return 1
difference = pla_error - svm_error
svm_better = helper(difference)
total_support_vectors = sum([ 1 for x in svm_weight if x >= 10*-3 ])
return svm_better, total_support_vectors
def trial(in_sample, out_of_sample):
target_function = random_target_function()
training_set = random_set(in_sample, target_function)
pla_weight = pla(training_set.z, training_set.y)
svm_weight = svm(training_set.z, training_set.y)
testing_set = random_set(out_of_sample, target_function)
pla_error, svm_error = [
weight_error(weight, testing_set.z, testing_set.y)
for weight in [pla_weight, svm_weight]
]
def helper(x):
if x > 0:
return 0
else:
return 1
difference = pla_error - svm_error
svm_better = helper(difference)
total_support_vectors = sum([1 for x in svm_weight if x >= 10 * -3])
return svm_better, total_support_vectors
def train_test(training_set, testing_set, learn, learn_args=[]):
weight = learn(training_set.z, training_set.y, *learn_args)
in_sample_error = weight_error(weight, training_set.z, training_set.y)
out_of_sample_error = weight_error(weight, testing_set.z, testing_set.y)
return [in_sample_error, out_of_sample_error]
def best_model(model_weights, testing_set):
errors = [ weight_error(
weight, testing_set.z[:,:len(weight)], testing_set.y)
for weight in model_weights ]
return len(model_weights[np.argmin(errors)]) - 1, errors # return k value that yields least error. see k_values
def train_test(training_set, testing_set, learn, learn_args=[]):
weight = learn(training_set.z, training_set.y, *learn_args)
in_sample_error = weight_error(weight, training_set.z, training_set.y)
out_of_sample_error = weight_error(weight, testing_set.z, testing_set.y)
return [ in_sample_error, out_of_sample_error ]
def trial(training_data, testing_data, a):
training_set = DataML(training_data, transform)
weights = minimize_error_aug(training_set.z, training_set.y, a)
in_error, out_error = [ weight_error(weights, tset.z, tset.y)
for tset in [training_set, DataML(testing_data, transform)] ]
return in_error, out_error