Parameter

class astropy.modeling.Parameter(name='', description='', default=None, unit=None, getter=None, setter=None, fixed=False, tied=False, min=None, max=None, bounds=None, prior=None, posterior=None, model=None)

Bases: astropy.utils.misc.OrderedDescriptor

Wraps individual parameters.

This class represents a model’s parameter (in a somewhat broad sense). It acts as both a descriptor that can be assigned to a class attribute to describe the parameters accepted by an individual model (this is called an “unbound parameter”), or it can act as a proxy for the parameter values on an individual model instance (called a “bound parameter”).

Parameter instances never store the actual value of the parameter directly. Rather, each instance of a model stores its own parameters parameter values in an array. A bound Parameter simply wraps the value in a Parameter proxy which provides some additional information about the parameter such as its constraints. In other words, this is a high-level interface to a model’s adjustable parameter values.

Unbound Parameters are not associated with any specific model instance, and are merely used by model classes to determine the names of their parameters and other information about each parameter such as their default values and default constraints.

See Parameters for more details.

Parameters:

name : str

parameter name

Warning

The fact that Parameter accepts name as an argument is an implementation detail, and should not be used directly. When defining a new Model class, parameter names are always automatically defined by the class attribute they’re assigned to.

description : str

parameter description

default : float or array

default value to use for this parameter

unit : Unit

if specified, the parameter will be in these units, and when the parameter is updated in future, it should be set to a Quantity that has equivalent units.

getter : callable

a function that wraps the raw (internal) value of the parameter when returning the value through the parameter proxy (eg. a parameter may be stored internally as radians but returned to the user as degrees)

setter : callable

a function that wraps any values assigned to this parameter; should be the inverse of getter

fixed : bool

if True the parameter is not varied during fitting

tied : callable or False

if callable is supplied it provides a way to link the value of this parameter to another parameter (or some other arbitrary function)

min : float

the lower bound of a parameter

max : float

the upper bound of a parameter

bounds : tuple

specify min and max as a single tuple–bounds may not be specified simultaneously with min or max

model : Model instance

binds the the Parameter instance to a specific model upon instantiation; this should only be used internally for creating bound Parameters, and should not be used for Parameter descriptors defined as class attributes

Attributes Summary

bounds	The minimum and maximum values of a parameter as a tuple
constraints	Types of constraints a parameter can have.
default	Parameter default value
fixed	Boolean indicating if the parameter is kept fixed during fitting.
max	A value used as an upper bound when fitting a parameter
min	A value used as a lower bound when fitting a parameter
name	Parameter name
posterior
prior
quantity	This parameter, as a Quantity instance.
shape	The shape of this parameter’s value array.
size	The size of this parameter’s value array.
tied	Indicates that this parameter is linked to another one.
unit	The unit attached to this parameter, if any.
validator	Used as a decorator to set the validator method for a Parameter.
value	The unadorned value proxied by this parameter.

Methods Summary

copy([name, description, default, unit, …])

Make a copy of this Parameter, overriding any of its core attributes in the process (or an exact copy).

Attributes Documentation

bounds

The minimum and maximum values of a parameter as a tuple

constraints = ('fixed', 'tied', 'bounds', 'prior', 'posterior')

Types of constraints a parameter can have. Excludes ‘min’ and ‘max’ which are just aliases for the first and second elements of the ‘bounds’ constraint (which is represented as a 2-tuple). ‘prior’ and ‘posterior’ are available for use by user fitters but are not used by any built-in fitters as of this writing.

default

Parameter default value

fixed

Boolean indicating if the parameter is kept fixed during fitting.

max

A value used as an upper bound when fitting a parameter

min

A value used as a lower bound when fitting a parameter

name

Parameter name

posterior

prior

quantity

This parameter, as a Quantity instance.

shape

The shape of this parameter’s value array.

size

The size of this parameter’s value array.

tied

Indicates that this parameter is linked to another one.

A callable which provides the relationship of the two parameters.

unit

The unit attached to this parameter, if any.

On unbound parameters (i.e. parameters accessed through the model class, rather than a model instance) this is the required/ default unit for the parameter.

validator

Used as a decorator to set the validator method for a Parameter. The validator method validates any value set for that parameter. It takes two arguments–self, which refers to the Model instance (remember, this is a method defined on a Model), and the value being set for this parameter. The validator method’s return value is ignored, but it may raise an exception if the value set on the parameter is invalid (typically an InputParameterError should be raised, though this is not currently a requirement).

The decorator returns the Parameter instance that the validator is set on, so the underlying validator method should have the same name as the Parameter itself (think of this as analogous to property.setter). For example:

>>> from astropy.modeling import Fittable1DModel
>>> class TestModel(Fittable1DModel):
...     a = Parameter()
...     b = Parameter()
...
...     @a.validator
...     def a(self, value):
...         # Remember, the value can be an array
...         if np.any(value < self.b):
...             raise InputParameterError(
...                 "parameter 'a' must be greater than or equal "
...                 "to parameter 'b'")
...
...     @staticmethod
...     def evaluate(x, a, b):
...         return a * x + b
...
>>> m = TestModel(a=1, b=2)  
Traceback (most recent call last):
...
InputParameterError: parameter 'a' must be greater than or equal
to parameter 'b'
>>> m = TestModel(a=2, b=2)
>>> m.a = 0  
Traceback (most recent call last):
...
InputParameterError: parameter 'a' must be greater than or equal
to parameter 'b'

On bound parameters this property returns the validator method itself, as a bound method on the Parameter. This is not often as useful, but it allows validating a parameter value without setting that parameter:

>>> m.a.validator(42)  # Passes
>>> m.a.validator(-42)  
Traceback (most recent call last):
...
InputParameterError: parameter 'a' must be greater than or equal
to parameter 'b'

value

The unadorned value proxied by this parameter.

Methods Documentation

copy(name=None, description=None, default=None, unit=None, getter=None, setter=None, fixed=False, tied=False, min=None, max=None, bounds=None, prior=None, posterior=None)

Make a copy of this Parameter, overriding any of its core attributes in the process (or an exact copy).

The arguments to this method are the same as those for the Parameter initializer. This simply returns a new Parameter instance with any or all of the attributes overridden, and so returns the equivalent of:

Parameter(self.name, self.description, ...)