from kernel_exp_family.estimators.estimator_oop import EstimatorBase
try:
from kernel_exp_family.estimators.parameter_search_bo import BayesOptSearch
except ImportError:
print("Could not import BayesOptSearch.")
from kernel_exp_family.kernels.kernels import gaussian_kernel, \
gaussian_kernel_grad, theano_available
from kernel_exp_family.tools.assertions import assert_array_shape
from kernel_exp_family.tools.log import Log
import numpy as np
if theano_available:
from kernel_exp_family.kernels.kernels import gaussian_kernel_hessian_theano, \
gaussian_kernel_third_order_derivative_tensor_theano
logger = Log.get_logger()
def compute_b(X, Y, K_XY, sigma):
assert X.shape[1] == Y.shape[1]
assert K_XY.shape[0] == X.shape[0]
assert K_XY.shape[1] == Y.shape[0]
NX = len(X)
D = X.shape[1]
b = np.zeros(NX)
K1 = np.sum(K_XY, 1)
for l in np.arange(D):
x_l = X[:, l]
y_l = Y[:, l]
from pybo import inits, recommenders, solvers, policies
import reggie
from reggie.means._core import Mean
from kernel_exp_family.tools.log import Log
import matplotlib.pyplot as plt
import numpy as np
logger = Log.get_logger()
class GPMean(Mean):
"""
Allows prescribing a posterior GP mean as the prior mean for
another GP.
Taken from dev-branch of reggie
"""
def __init__(self, gp):
super(GPMean, self).__init__()
self._gp = gp.copy()
def __info__(self):
info = []
info.append(('gp', self._gp))
return info
def get_mean(self, X):
mu, _ = self._gp.predict(X)
return mu