def train(self, x, y):
if len(y.shape) == 1:
y = y.reshape((y.size, 1))
elif len(y.shape) == 2:
if y.shape[1] != 1:
# If y is provided as a row instead of a column,
# just transpose it into a column
if y.shape[0] == 1:
y = y.T
else:
raise ValueError('all training results must have one column')
else:
raise ValueError('y must be a 1- or 2-dimensional matrix')
if x.shape[0] != y.shape[0]:
raise ValueError('you must provide the same number of input features as input results')
if self.normalizer:
self.normalizer.set_basis(x)
x_train = self.normalizer.normalize(x) if self.normalizer else x
if self.pad_x:
x_train = pad_ones(x_train)
self.theta = self.generator.calculate(x_train, y)
self.theta = self.theta.reshape([self.theta.shape[0], 1])
def predict(self, x):
if not self.theta.size:
raise Exception('this classifier has not been trained yet')
if self.normalizer:
x = self.normalizer.normalize(x)
n = x.shape[1]
expected_n = self.theta.shape[0] - (1 if self.pad_x else 0)
if n != expected_n:
raise ValueError('this classifier was trained using inputs with %s features but this input has %s features'
% (str(expected_n), str(n)))
x_predict = pad_ones(x) if self.pad_x else x
return x_predict.dot(self.theta)
