def set_samplers(self, gp_model):
"""
Defines the samplers of the parameters of the model.
We assume that we only have one set of length scale parameters.
"""
gp_model.slice_samplers = []
gp_model.samples_parameters = []
gp_model.start_point_sampler = []
ignore_index = None
if not self.parametric_mean:
slice_parameters = {
'max_steps_out': gp_model.max_steps_out,
'component_wise': True,
}
else:
slice_parameters = {
'max_steps_out': gp_model.max_steps_out,
'component_wise': False,
}
dimension_sampler = gp_model.dimension_parameters
if self.parametric_mean:
dimension_sampler += self.parametrics.n_weights + self.parametrics.n_parameters
gp_model.slice_samplers.append(SliceSampling(
self.log_prob, range(dimension_sampler), ignore_index=ignore_index,
**slice_parameters))
if gp_model.start_point_sampler is not None and len(gp_model.start_point_sampler) > 0:
if len(gp_model.samples_parameters) == 0:
gp_model.samples_parameters.append(np.array(gp_model.start_point_sampler))
else:
gp_model.samples_parameters = []
z = gp_model.get_value_parameters_model
z = z[2:]
if self.parametric_mean:
if gp_model.mean_params is None or len(gp_model.mean_params) == 0:
gp_model.mean_params = self.fit_parametric_functions(gp_model.data['evaluations'])
mean_params = gp_model.mean_params
gp_model.samples_parameters.append(np.concatenate((z, mean_params)))
else:
gp_model.samples_parameters.append(z)
if gp_model.n_burning > 0:
parameters = self.sample_parameters(
gp_model, float(gp_model.n_burning) / (gp_model.thinning + 1))
gp_model.samples_parameters = []
gp_model.samples_parameters.append(parameters[-1])
gp_model.start_point_sampler = parameters[-1]
else:
gp_model.start_point_sampler = gp_model.samples_parameters[-1]
def set_samplers(self):
gp_model = self.gp_model
gp_model.slice_samplers = []
gp_model.samples_parameters = []
gp_model.start_point_sampler = []
ignore_index = None
slice_parameters = {
'max_steps_out': gp_model.max_steps_out,
'component_wise': True,
}
dimension_sampler = gp_model.dimension_parameters
gp_model.slice_samplers.append(SliceSampling(
self.log_prob, range(dimension_sampler), ignore_index=ignore_index,
**slice_parameters))
if gp_model.start_point_sampler is not None and len(gp_model.start_point_sampler) > 0:
if len(gp_model.samples_parameters) == 0:
gp_model.samples_parameters.append(np.array(gp_model.start_point_sampler))
else:
gp_model.samples_parameters = []
self.samples_parameters = []
z = gp_model.get_value_parameters_model
z = z[2:]
gp_model.samples_parameters.append(z)
self.samples_parameters.append(z)
if gp_model.n_burning > 0 and self.burning:
parameters = self.sample_parameters(
gp_model, float(gp_model.n_burning) / (gp_model.thinning + 1))
self.samples_parameters = []
self.samples_parameters.append(parameters[-1])
self.start_point_sampler = parameters[-1]
gp_model.samples_parameters = []
gp_model.samples_parameters.append(parameters[-1])
gp_model.start_point_sampler = parameters[-1]
else:
gp_model.start_point_sampler = gp_model.samples_parameters[-1]
self.start_point_sampler = gp_model.samples_parameters[-1]
def set_samplers(self):
self.slice_samplers = []
self.samples_parameters = []
self.start_point_sampler = []
slice_parameters = {
'max_steps_out': 1000,
'component_wise': False,
}
n_params = self.parametrics.n_parameters + self.parametrics.n_weights + 1
sampler = SliceSampling(self.log_prob, range(0, len(n_params)),
**slice_parameters)
self.slice_samplers.append(sampler)
if self.start_point_sampler is not None and len(
self.start_point_sampler) > 0:
if len(self.samples_parameters) == 0:
self.samples_parameters.append(
np.array(self.start_point_sampler))
else:
self.samples_parameters = []
start_point = self.fit_parametric_functions(
self.training_data['evaluations'])
self.samples_parameters.append(start_point)
if self.n_burning > 0:
parameters = self.sample_parameters(
self.slice_samplers[0],
float(self.n_burning) / (self.thinning + 1))
self.samples_parameters = []
self.samples_parameters.append(parameters[-1])
self.start_point_sampler = parameters[-1]
else:
self.start_point_sampler = self.samples_parameters[-1]
def set_samplers(self):
slice_parameters = {
'max_steps_out': 1000,
'component_wise': False,
}
n_params = self.parametrics.n_weights + self.parametrics.n_parameters + 1
sampler = SliceSampling(self.log_prob_slice, range(0, n_params),
**slice_parameters)
self.sampler = sampler
start = self.parametrics.get_starting_values_mle(self.training_data)
samples_parameters = []
samples_parameters.append(start)
if self.n_burning > 0:
parameters = self.sample_parameters(
float(self.n_burning) / (self.n_thinning + 1), start, sampler,
self.training_data)
samples_parameters = []
samples_parameters.append(parameters[-1])
self.samples_parameters = samples_parameters