from pynolca.nolca import NOLCA
from pynolca import kernel
from pynolca import loss
from pynolca import preprocessing
import numpy as np
np.random.seed(2019)
# Load data and split
data = datasets.load_digits()
X, y = preprocessing.shuffle(data.data, data.target)
num_stage_1 = len(y) // 2
X_stage_1, y_stage_1 = X[: num_stage_1], y[: num_stage_1]
X_stage_2, y_stage_2 = X[num_stage_1: ], y[num_stage_1: ]
# Construct kernel and loss
demo_kernel = kernel.Polynomial_Kernel(scale_factor = 1, intercept = 1,
degree = 3)
demo_loss = loss.Ramp(parameter = -0.1, policy = "static")
# Construct the classifier
clf = NOLCA(demo_kernel, demo_loss)
# Train the classifier
clf.training(X_stage_1[: -1], y_stage_1[: -1], learning_rate = 0.01,
reg_coefficient = 0)
# Visualization
clf.plot_accuracy_curve()
clf.plot_confusion_matrix()
# Predict
print clf.predicting(X_stage_1[-1])
# Retrain
clf_new = NOLCA(demo_kernel, demo_loss,
num_support_vectors = clf.get_num_support_vectors(),
support_vectors = clf.get_support_vectors(),
from pynolca import loss
from pynolca import preprocessing
import numpy as np
from time import time
data = datasets.load_digits()
X, y = data.data, data.target
print(X.shape)
accuracies = []
num_support_vectors = []
running_time = []
parameters = [-1, -10**-13, -10**-14, -10**-15]
for i in range(len(parameters)):
# Construct kernel and loss
Kernel = kernel.RBF_Kernel()
Loss = loss.Ramp(policy="static", parameter=parameters[i])
# Construct classifier
clf = NOLCA(Kernel, Loss)
# Train
start = time()
clf.training(X, y, learning_rate=0.01)
end = time()
elapsed = end - start
# Visualization
num_support_vectors.append(clf.get_num_support_vectors())
accuracies.append(clf.get_accuracy()[-1])
running_time.append(elapsed)
print accuracies
print num_support_vectors
print running_time