def build_param(index):
"""Get one of the parameters of the Kurtosis matrix as an object.
Args:
tuple: index, 4 integers specyfing the location of this parameter in the matrix.
Returns:
Parameter: a constructed parameter to be used in a compartment model.
"""
if len(set(index)) == 1:
_lower_bound = 0
_init_value = 0.001
_sampling_prior = UniformWithinBoundsPrior()
else:
_lower_bound = -1e20
_init_value = 0
_sampling_prior = AlwaysOne()
class matrix_element_param(FreeParameterTemplate):
name = 'W_{}{}{}{}'.format(*index)
init_value = _init_value
lower_bound = _lower_bound
upper_bound = 1e20
parameter_transform = ScaleTransform(100)
sampling_prior = _sampling_prior
sampling_proposal_std = 0.01
return ParameterBuilder().create_class(matrix_element_param)()
class w(FreeParameterTemplate):
init_value = 0.5
lower_bound = 0
upper_bound = 1
sampling_proposal_std = 0.01
sampling_prior = UniformWithinBoundsPrior()
numdiff_info = {'scale_factor': 10}
class FreeParameterTemplate(ParameterTemplate):
"""The default template options for free parameters.
Attributes:
init_value (float): the initial value
fixed (boolean or ndarray of float): if this parameter is fixed or not. If not fixed this should
hold a reference to a value or a matrix
lower_bound (float): the lower bounds
upper_bound (float): the upper bounds
parameter_transform
(str or :class:`~mdt.model_building.parameter_functions.transformations.AbstractTransformation`): the
parameter transformation, this is used for automatic range transformation of the parameters during
optimization. See Harms 2017 NeuroImage for details. Typical elements are:
* ``Identity``: no transformation
* ``Positivity``: ensures the parameters are positive
* ``Clamp``: limits the parameter between its lower and upper bounds
* ``CosSqrClamp``: changes the range of the optimized parameters to [0, 1] and ensures boundary constraints
* ``SinSqrClamp``: same as ``CosSqrClamp``
* ``SqrClamp``: same as clamp but with an additional square root to change the magnitude of the range
* ``AbsModPi``: ensures absolute modulus of the input parameters between zero and pi.
* ``AbsModTwoPi``: ensures absolute modulus of the input parameters between zero and two pi.
sampling_proposal_std (float): the default proposal standard deviation for this parameter. This is used
in some MCMC sampling routines.
sampling_proposal_modulus (float or None): if given, a modulus we will use when finalizing the proposal
distributions. That is, when we are finalizing the proposals we will take, if set, the absolute
modulus of that parameter to ensure the parameter is within [0, <modulus>].
sampling_prior: the prior function
numdiff_info (dict or :class:`~mdt.model_building.parameter_functions.numdiff_info.NumDiffInfo`):
the information necessary to take the numerical derivative of a model with respect to this parameter.
Either a dictionary with the keyword arguments to
:class:`~mdt.model_building.parameter_functions.numdiff_info.SimpleNumDiffInfo` or an information
object directly. If None, we use an empty dictionary. Please note that if you override this, you will have
to specify all of the items (no automatic inheritance of sub-items).
"""
data_type = 'mot_float_type'
fixed = False
init_value = 0.03
lower_bound = 0.0
upper_bound = 4.0
parameter_transform = 'Identity'
sampling_proposal_std = 1
sampling_proposal_modulus = None
sampling_prior = UniformWithinBoundsPrior()
numdiff_info = {
'max_step': 0.1,
'scale_factor': 1,
'use_bounds': True,
'modulus': None,
'use_upper_bound': True,
'use_lower_bound': True
}