def run_multiclass():
from data_loader import toy_data_multiclass_3_classes_non_separable, \
toy_data_multiclass_5_classes, \
data_loader_mnist
datasets = [(toy_data_multiclass_3_classes_non_separable(),
'Synthetic data', 3),
(toy_data_multiclass_5_classes(), 'Synthetic data', 5),
(data_loader_mnist(), 'MNIST', 10)]
for data, name, num_classes in datasets:
print('%s: %d class classification' % (name, num_classes))
X_train, X_test, y_train, y_test = data
print('One-versus-rest:')
w, b = OVR_train(X_train, y_train, C=num_classes)
train_preds = OVR_predict(X_train, w=w, b=b)
preds = OVR_predict(X_test, w=w, b=b)
print('train acc: %f, test acc: %f' % (accuracy_score(
y_train, train_preds), accuracy_score(y_test, preds)))
print('Multinomial:')
w, b = multinomial_train(X_train, y_train, C=num_classes)
train_preds = multinomial_predict(X_train, w=w, b=b)
preds = multinomial_predict(X_test, w=w, b=b)
print('train acc: %f, test acc: %f' % (accuracy_score(
y_train, train_preds), accuracy_score(y_test, preds)))
def run_binary():
from data_loader import toy_data_binary, \
data_loader_mnist
print('Performing binary classification on synthetic data')
X_train, X_test, y_train, y_test = toy_data_binary()
w, b = binary_train(X_train, y_train)
train_preds = binary_predict(X_train, w, b)
preds = binary_predict(X_test, w, b)
print(
'train acc: %f, test acc: %f' %
(accuracy_score(y_train, train_preds), accuracy_score(y_test, preds)))
print('Performing binary classification on binarized MNIST')
X_train, X_test, y_train, y_test = data_loader_mnist()
binarized_y_train = [0 if yi < 5 else 1 for yi in y_train]
binarized_y_test = [0 if yi < 5 else 1 for yi in y_test]
w, b = binary_train(X_train, binarized_y_train)
train_preds = binary_predict(X_train, w, b)
preds = binary_predict(X_test, w, b)
print('train acc: %f, test acc: %f' %
(accuracy_score(binarized_y_train, train_preds),
accuracy_score(binarized_y_test, preds)))
def run_multiclass():
from data_loader import smile_dataset_clear, \
smile_dataset_blur, \
data_loader_mnist
import time
datasets = [(smile_dataset_clear(), 'Clear smile data', 3),
(smile_dataset_blur(), 'Blur smile data', 3),
(data_loader_mnist(), 'MNIST', 10)]
for data, name, num_classes in datasets:
print('%s: %d class classification' % (name, num_classes))
X_train, X_test, y_train, y_test = data
for gd_type in ["sgd", "gd"]:
s = time.time()
w, b = sol.multiclass_train(X_train,
y_train,
C=num_classes,
gd_type=gd_type)
print(gd_type + ' training time: %0.6f seconds' %
(time.time() - s))
train_preds = sol.multiclass_predict(X_train, w=w, b=b)
preds = sol.multiclass_predict(X_test, w=w, b=b)
print('train acc: %f, test acc: %f' % (accuracy_score(
y_train, train_preds), accuracy_score(y_test, preds)))
print()
def run_binary():
from data_loader import toy_data_binary, \
moon_dataset, \
data_loader_mnist
datasets = [(toy_data_binary(), 'Synthetic data'),
(moon_dataset(), 'Two Moon data'),
(data_loader_mnist(), 'Binarized MNIST data')]
for data, name in datasets:
print(name)
X_train, X_test, y_train, y_test = toy_data_binary()
if name == 'Binarized MNIST data':
y_train = [0 if yi < 5 else 1 for yi in y_train]
y_test = [0 if yi < 5 else 1 for yi in y_test]
y_train = np.asarray(y_train)
y_test = np.asarray(y_test)
for loss_type in ["perceptron", "logistic"]:
w, b = sol.binary_train(X_train, y_train, loss="logistic")
train_preds = sol.binary_predict(X_train, w, b, loss=loss_type)
preds = sol.binary_predict(X_test, w, b, loss=loss_type)
print(loss_type + ' train acc: %f, test acc: %f' %
(accuracy_score(y_train, train_preds),
accuracy_score(y_test, preds)))
print()
from data_loader import smile_dataset_clear, \
smile_dataset_blur, \
data_loader_mnist
import time
datasets = [(smile_dataset_clear(), 'Clear smile data', 3),
(smile_dataset_blur(), 'Blur smile data', 3),
(data_loader_mnist(), 'MNIST', 10)]
for data, name, num_classes in datasets:
print('%s: %d class classification' % (name, num_classes))
X_train, X_test, y_train, y_test = data