def fit(self, data, labels, update=True):
'''
Calculates the gradient and train loss.
If the update flag is set to False, gradient is calculated but own weights will not be updated
'''
total_delta_w = {}
train_loss = 0
for x, label in zip(data, labels):
xw = dotproduct(x, self.__w)
if self.__misclassification(xw, label):
delta_w = self.__gradient(x, label)
if update:
self.update_weights(delta_w)
else:
delta_w = self.__regularization_gradient(x)
if update:
self.update_weights(delta_w)
for k, v in delta_w.items():
if k in total_delta_w:
total_delta_w[k] += v
else:
total_delta_w[k] = v
train_loss += max(1 - label * xw, 0)
train_loss += self.__regularizer(x)
return total_delta_w, train_loss / len(labels)
def loss(self, data, labels):
''' Returns the MSE loss of the data with the true labels. '''
total_loss = 0
for x, label in zip(data, labels):
xw = dotproduct(x, self.__w)
total_loss += max(1 - label * xw, 0)
total_loss += self.__regularizer(x)
return total_loss / len(labels)
def predict(self, data):
''' Predict the labels of the input data '''
return [sign(dotproduct(x, self.__w)) for x in data]