class BayesianRegression(object):
"""Bayesian linear regression."""
def __init__(self,
n_feature,
prior_mean=0,
prior_precision=1e-6,
prior_a=10,
prior_b=1):
super().__init__()
self.n_feature = n_feature
if np.shape(prior_mean) != (n_feature, ):
prior_mean = prior_mean * np.ones(n_feature)
if np.shape(prior_precision) != (n_feature, n_feature):
prior_precision = prior_precision * np.ones(n_feature)
self.prior_mean = prior_mean
self.prior_precision = prior_precision
self.prior_a = prior_a
self.prior_b = prior_b
self._init_weights()
# print("Intialize regression")
# self.print()
def _init_weights(self):
self.weights = GaussianARD(self.prior_mean,
self.prior_precision,
shape=(self.n_feature, ))
def fit(self, X, y):
self._init_weights()
# self.cost,
# self.myopic_voc(action, state),
# self.vpi_action(action, state),
# self.vpi(state),
# self.expected_term_reward(state)
self.tau = Gamma(self.prior_a, self.prior_b)
F = SumMultiply('i,i', self.weights, X)
y_obs = GaussianARD(F, self.tau)
y_obs.observe(y)
Q = VB(y_obs, self.weights)
Q.update(repeat=10, tol=1e-4, verbose=False)
def predict(self, x, return_var=False):
y = SumMultiply('i,i', self.weights, x)
y_hat, var, *_ = y.get_moments()
if return_var:
return y_hat, var
else:
return y_hat
def sample(self, x):
w = self.weights.random()
return x @ w
def print(self, diagonal=True):
mean, m2 = self.weights.get_moments()[:2]
var = m2 - mean**2
if diagonal:
var = np.diagonal(var)
bar = '_' * 40
print(f'{bar}\n{mean.round(3)}\n{var.round(3)}\n{bar}')
