def predict(self, x):
"""
Predict the output given an input array
:param x: input data
:return: predicted values
"""
pol = pps.pps_basis(x, self.model['order'])
return np.dot(pol, self.model['weights'])
def train(self, x, y, l2_term=0, method='closed'):
"""
Train the PPS Regression model
:param x: input data
:param y: output data
:param l2_term: regularization term
:param method: optimization method
"""
# train using the closed form
if method == 'closed':
# generate the sigmoid polynomial
H = pps.pps_basis(x, self.model['order'])
H_t = np.transpose(H)
# apply the regularization term
regularizer = np.identity(self.model['order'] + 1) * l2_term
A = linalg.inv(np.dot(H_t, H) + regularizer)
# compute the best parameters
self.model['weights'] = np.dot(np.dot(A, H_t), y)
