def example2():
my_random_generator = random.Random()
my_random_generator.seed(0)
# dense moons data set
X_train_l, L_train_l, X_train_u, X_test, L_test = examples.get_moons_data(
my_random_generator)
X_test = np.array(X_test)
L_train_l = (np.array(L_train_l) + 1) // 2
L_test = (np.array(L_test) + 1) // 2
print("Unique labels: ", np.unique(L_test))
X_train = np.concatenate((X_train_l, X_train_u), axis=0)
y_train = np.hstack((L_train_l, -np.ones(len(X_train_u))))
X1 = X_train[:, 0].reshape((-1, 1))
X2 = X_train[:, 1].reshape((-1, 1))
print('Logistic')
base_lr = LogisticRegression()
base_lr.fit(X_train_l, L_train_l)
#base_lr.fit(X_test, L_test)
y_pred = base_lr.predict(X_test)
print(classification_report(L_test, y_pred))
print('Logistic CoTraining')
t_start = time.time()
lg_co_clf = CoTrainingClassifier(LogisticRegression())
lg_co_clf.fit(X1, X2, y_train)
t_end = time.time()
y_pred_train = lg_co_clf.predict(X_train[:, 0].reshape((-1, 1)),
X_train[:, 1].reshape((-1, 1)))
y_pred = lg_co_clf.predict(X_test[:, 0].reshape((-1, 1)),
X_test[:, 1].reshape((-1, 1)))
print(classification_report(L_test, y_pred))
return
def s3vm_moon():
my_random_generator = random.Random()
my_random_generator.seed(0)
# dense moons data set
X_train_l, L_train_l, X_train_u, X_test, L_test = examples.get_moons_data(
my_random_generator)
print("Unique labels: ", np.unique(L_test))
print("X_train_l: ", X_train_l)
print("L_train_l: ", L_train_l)
print("X_train_u: ", X_train_u)
print("X_test: ", X_test)
print("L_test: ", L_test)
t_start = time.time()
# the parameter estimate_r for the balance ratio has to be provided explicitly, since the estimation
# is bad due to only 5 labeled patterns in the training set.
model = QN_S3VM(
X_train_l,
L_train_l,
X_train_u,
my_random_generator,
lam=0.0009765625,
lamU=1,
kernel_type="RBF",
sigma=0.5,
estimate_r=0.0,
)
preds_train = model.train()
t_end = time.time()
elapsed_time = t_end - t_start
preds = model.getPredictions(X_test)
error = examples.classification_error(preds, L_test)
print("Time needed to compute the model: ", elapsed_time, " seconds")
print("Classification error of QN-S3VM: ", error)
X_draw = np.concatenate((X_train_l, X_train_u), axis=0)
y_draw = np.hstack((L_train_l, np.zeros(len(X_train_u))))
disp(X_draw, y_draw)
disp(X_draw, preds_train)
disp(np.array(X_test), np.array(preds))
return
model = QN_S3VM(X_train_l,
L_train_l,
X_train_u,
my_random_generator,
lam=1,
lamU=1)
model.train()
t_end = time.time()
elapsed_time = t_end - t_start
preds = model.getPredictions(X_test)
error = examples.classification_error(preds, L_test)
print("Time needed to compute the model: ", elapsed_time, " seconds")
print("Classification error of QN-S3VM: ", error)
# dense moons data set
X_train_l, L_train_l, X_train_u, X_test, L_test = examples.get_moons_data(
my_random_generator)
t_start = time.time()
# the parameter estimate_r for the balance ratio has to be provided explicitly, since the estimation
# is bad due to only 5 labeled patterns in the training set.
model = QN_S3VM(X_train_l, L_train_l, X_train_u, my_random_generator, lam=0.0009765625, lamU=1, \
kernel_type="RBF", sigma=0.5, estimate_r=0.0,)
model.train()
t_end = time.time()
elapsed_time = t_end - t_start
preds = model.getPredictions(X_test)
error = examples.classification_error(preds, L_test)
print("Time needed to compute the model: ", elapsed_time, " seconds")
print("Classification error of QN-S3VM: ", error)
print("")
print "Classification error of QN-S3VM: ", error
# dense gaussian data set
X_train_l, L_train_l, X_train_u, X_test, L_test = examples.get_gaussian_data(my_random_generator)
t_start = time.time()
model = QN_S3VM(X_train_l, L_train_l, X_train_u, my_random_generator, lam=1, lamU=1)
model.train()
t_end = time.time()
elapsed_time = t_end - t_start
preds = model.getPredictions(X_test)
error = examples.classification_error(preds,L_test)
print "Time needed to compute the model: ", elapsed_time, " seconds"
print "Classification error of QN-S3VM: ", error
# dense moons data set
X_train_l, L_train_l, X_train_u, X_test, L_test = examples.get_moons_data(my_random_generator)
t_start = time.time()
# the parameter estimate_r for the balance ratio has to be provided explicitly, since the estimation
# is bad due to only 5 labeled patterns in the training set.
model = QN_S3VM(X_train_l, L_train_l, X_train_u, my_random_generator, lam=0.0009765625, lamU=1, \
kernel_type="RBF", sigma=0.5, estimate_r=0.0,)
model.train()
t_end = time.time()
elapsed_time = t_end - t_start
preds = model.getPredictions(X_test)
error = examples.classification_error(preds,L_test)
print "Time needed to compute the model: ", elapsed_time, " seconds"
print "Classification error of QN-S3VM: ", error
print ""