autofit.ModelMapper#

class autofit.ModelMapper(*arguments, **kwargs)[source]#

Bases: Collection

A mapper of priors formed by passing in classes to be reconstructed

@DynamicAttrs

The ModelMapper converts a set of classes whose input attributes may be modeled using a non-linear search, to parameters with priors attached.

A config is passed into the model mapper to provide default setup values for the priors:

mapper = ModelMapper(config)

All class instances that are to be generated by the model mapper are specified by adding classes to it:

mapper = ModelMapper()

mapper.sersic = al.lp.AbstractSersic mapper.gaussian = al.lp.Gaussian mapper.any_class = SomeClass

A Model instance is created each time we add a class to the mapper. We can access those models using # the mapper attributes:

sersic_model = mapper.sersic

This allows us to replace the default priors:

mapper.sersic.normalization = GaussianPrior(mean=2., sigma=5.)

Or maybe we want to tie two priors together:

mapper.sersic.two = mapper.other_sersic.two

This statement reduces the number of priors by one and means that the two sersic instances will always share # the same rotation two two.

We can then create instances of every class for a unit hypercube vector with length equal to # len(mapper.priors):

model_instance = mapper.model_instance_for_vector([.4, .2, .3, .1])

The attributes of the model_instance are named the same as those of the mapper:

sersic_1 = mapper.sersic_1

But this attribute is an instance of the actual AbstractSersic:P class

A ModelMapper can be concisely constructed using keyword arguments:

mapper = prior.ModelMapper(: source_light_profile=light_profile.AbstractSersic, lens_mass_profile=mass_profile.IsothermalCore, lens_light_profile=light_profile.SersicCore

)

The object multiple Python classes are input into to create model-components, which has free parameters that are fitted by a non-linear search.

Multiple Python classes can be input into a Collection in order to compose high dimensional models made of multiple model-components.

The Collection object is highly flexible, and can create models from many input Python data structures (e.g. a list of classes, dictionary of classes, hierarchy of classes).

For a complete description of the model composition API, see the PyAutoFit model API cookbooks:

https://pyautofit.readthedocs.io/en/latest/cookbooks/cookbook_1_basics.html

The Python class input into a Model to create a model component is written using the following format:

The name of the class is the name of the model component (e.g. Gaussian).
The input arguments of the constructor are the parameters of the mode (e.g. centre, normalization and sigma).
The default values of the input arguments tell PyAutoFit whether a parameter is a single-valued float or a

multi-valued tuple.

[Rich document more clearly]

A prior model used to represent a list of prior models for convenience.

Arguments are flexibly converted into a collection.

Parameters:: arguments – Classes, prior models, instances or priors

Examples

class Gaussian:

def __init__(
self, centre=0.0, # <- PyAutoFit recognises these normalization=0.1, # <- constructor arguments are sigma=0.01, # <- the Gaussian’s parameters.

):
self.centre = centre self.normalization = normalization self.sigma = sigma

model = af.Collection(gaussian_0=Gaussian, gaussian_1=Gaussian)

Methods

`add_assertion`	Assert that some relationship holds between physical values associated with priors at the point an instance is created.
`add_dict_items`
`all_paths_for_prior`	Find all paths that points at the given tuple.
`append`
`as_model`
`attribute_tuples_with_type`	Tuples describing the name and instance for attributes in the model with a given type, recursively.
`cast`	Cast models to a new type.
`check_assertions`	Check that all assertions are satisfied by the given arguments.
`copy`	Create a copy of the model.
`copy_with_fixed_priors`	Recursively overwrite priors in the mapper with instance values from the instance except where the containing class is the descendant of a listed class.
`dict`	A dictionary representation of this object
`direct_tuples_with_type`
`freeze`	Freeze this object.
`from_dict`	Recursively parse a dictionary returning the model, collection or instance that is represents.
`from_instance`	Recursively create a prior object model from an object model.
`from_json`	Loads the model from a .json file, which was written using the model's dictionary (dict) attribute as follows:
`from_object`
`gaussian_prior_model_for_arguments`	Create a new collection, updating its priors according to the argument dictionary.
`has`	param cls: The type to check for
`has_instance`	True iff this model contains an instance of type cls, recursively.
`has_model`	True iff this model contains a Model of type cls, recursively.
`index`	Retrieve the index of a given path in the model
`instance_for_arguments`	Returns an instance of the model for a set of arguments
`instance_from_path_arguments`	Create an instance from a dictionary mapping paths to tuples to corresponding values.
`instance_from_prior_medians`	Returns a list of physical values from the median values of the priors.
`instance_from_prior_name_arguments`	Instantiate the model from the names of priors and corresponding values.
`instance_from_unit_vector`	Returns a ModelInstance, which has an attribute and class instance corresponding to every Model attributed to this instance.
`instance_from_vector`	Returns a ModelInstance, which has an attribute and class instance corresponding to every Model attributed to this instance.
`is_only_model`	True iff this model contains at least one Model of type cls and contains no PriorModels that are not of type cls, recursively.
`items`
`log_prior_list_from`
`log_prior_list_from_vector`	Compute the log priors of every parameter in a vector, using the Prior of every parameter.
`mapper_from_partial_prior_arguments`	Returns a new model mapper from a dictionary mapping existing priors to new priors, keeping existing priors where no mapping is provided.
`mapper_from_prior_arguments`	Returns a new model mapper from a dictionary mapping existing priors to new priors.
`mapper_from_prior_means`	The widths of the new priors are taken from the width_config. The new gaussian priors must be provided in the same order as the priors associated with model. If a is not None then all priors are created with an absolute width of a. If r is not None then all priors are created with a relative width of r. :param means: The median PDF value of every Gaussian, which centres each GaussianPrior. :param no_limits: If True generated priors have infinite limits :param r: The relative width to be assigned to gaussian priors :param a: print(tuples[i][1], width) The absolute width to be assigned to gaussian priors :param tuples: A list of tuples each containing the mean and width of a prior.
`mapper_from_uniform_floats`	The widths of the new priors are the floats value minus and plus the input bound b.
`model_tuples_with_type`	All models of the class in this model which have at least one free parameter, recursively.
`models_with_type`	Return all models of a given type in the model tree.
`name_for_prior`	Construct a name for the prior.
`next_id`
`object_for_path`	Get the object at a given path.
`path_for_name`	Find the path to a prior in the model that matches a given name.
`path_for_prior`	Find a path that points at the given tuple.
`path_instance_tuples_for_class`	Tuples containing the path tuple and instance for every instance of the class in the model tree.
`prior_with_id`
`product`	Combine multiple models with the same structure by replacing priors with priors that contain a message which is the product of the messages of the priors with the same path in each model.
`random_instance`	Returns a random instance of the model.
`random_instance_from_priors_within_limits`	Returns a random instance of physical values by drawing uniform random values between lower and upper limits defined as unit values, using the model priors to map them from unit values to physical values.
`random_unit_vector_within_limits`	Generate a random vector of unit values by drawing uniform random values between 0 and 1.
`random_vector_from_priors_within_limits`	Returns a random vector of physical values by drawing uniform random values between lower and upper limits defined as unit values, using the model priors to map them from unit values to physical values.
`remove`
`replacing`
`replacing_for_path`	Create a new model replacing the value for a given path with a new value
`sort_priors_alphabetically`	Sort priors by their paths according to this model.
`take_attributes`	Take all attributes with a matching path from the source prior model.
`tree_flatten`
`tree_unflatten`
`unfreeze`	Unfreeze this object.
`vector_from_unit_vector`	param unit_vector: A unit hypercube vector
`with_limits`	Create a new instance of this model where each prior is updated to lie between new limits.
`with_paths`	Recursively generate a copy of this model retaining only objects specified by the list of paths.
`with_prefix`	Filter members of the collection, only returning those that start with a given prefix as a new collection.
`without_attributes`	Returns a copy of this object with all priors, prior models and constants removed.
`without_paths`	Recursively generate a copy of this model retaining only objects not specified by the list of paths.

Attributes

`all_name_prior_tuples`	Maps a tuple containing all names for a given prior to that prior.
`all_names`	All possible names for all priors grouped such that all names for a given prior are collected together in a tuple.
`all_paths`	All possible paths to all priors grouped such that all paths to the same prior are collected together in a tuple.
`all_paths_prior_tuples`	Maps a tuple containing all paths to a given prior to that prior.
`assertions`
`component_number`
`composition`
`direct_deferred_tuples`
`direct_instance_tuples`
`direct_prior_model_tuples`
`direct_prior_tuples`	returns: direct_priors :rtype: [(String, Prior)]
`direct_tuple_priors`
`identifier`
`info`	Use the priors that make up the model_mapper to generate information on each parameter of the overall model.
`instance_tuples`	returns: instances :rtype: [(String, instance)]
`joined_paths`
`label`
`mean_field`	Implements the same interface as the graphical code
`model_component_and_parameter_names`	Lists each parameter's name and path, and is used for labeling visualization with parameter labels.
`name`
`order_no`	A string that can be used to order models by their parametrisation.
`parameter_labels`	Returns a list of labels containing latex labels of every parameter in a model.
`parameter_labels_with_superscripts`	Returns a list of the latex parameter label and superscript of every parameter in a model.
`parameter_labels_with_superscripts_latex`	Returns a list of the latex parameter label and superscript of every parameter in a model.
`parameter_names`	Returns a list of labels containing the name of every parameter in a model.
`parameterization`	Describes the path to each of the PriorModels, its class and its number of free parameters
`path_float_tuples`
`path_priors_tuples`
`paths`	A list of paths to all the priors in the model, ordered by their ids
`physical_values_from_prior_medians`	returns: physical_values -- A list of physical values constructed by taking the mean possible value from
`prior_class_dict`
`prior_count`	How many unique priors does this model contain?
`prior_model_tuples`
`prior_prior_model_dict`	returns: prior_prior_model_dict -- A dictionary mapping priors to associated prior models. Each prior will only
`prior_tuples`	returns: priors :rtype: [(String, Prior))]
`prior_tuples_ordered_by_id`	returns: priors -- An ordered list of unique priors associated with this mapper :rtype: [Prior]
`priors`
`priors_ordered_by_id`
`random_vector_from_priors`	Generate a random vector of physical values by drawing uniform random values between 0 and 1 and using the model priors to map them from unit values to physical values.
`superscripts`	Returns a list of the model component superscripts for every parameter in a model.
`superscripts_overwrite_via_config`	Returns a list of the model component superscripts for every parameter in a model, which can be used to overwrite the default superscripts used in the function above.
`total_free_parameters`	Returns the prior count, but with a name that is more easy to interpret for users.
`tuple_prior_tuples`	returns: tuple_prior_tuples :rtype: [(String, TuplePrior)]
`unique_path_prior_tuples`
`unique_prior_paths`
`unique_prior_tuples`	returns: prior_tuple_dict -- The set of all priors associated with this mapper :rtype: [(Prior, PriorTuple)]
`values`

property prior_prior_model_dict#

returns: prior_prior_model_dict – A dictionary mapping priors to associated prior models. Each prior will only: have one prior model; if a prior is shared by two prior models then one of those prior models will be in this dictionary.

Return type:: {Prior: Model}