Source code notes on SubhaloModelFactory

This section of the documentation provides detailed notes on the source code implementation of the SubhaloModelFactory class. The purpose of the SubhaloModelFactory class is to provide a flexible, standardized platform for building subhalo-based models that can directly populate simulations with mock galaxies. The goal is to make it easy to swap new modeling features in and out of the framework while maintaining a uniform syntax. This way, when you want to study one particular feature of the galaxy-halo connection, you can focus exclusively on developing that feature, leaving the factory to take care of the remaining aspects of the mock population. This tutorial describes in detail how the SubhaloModelFactory accomplishes this standardization.


We will start in Overview of the factory design with a high-level description of how the class creates a composite model from a set of independently-defined features. In Inferring a model dictionary from the constructor inputs we describe how the factory’s __init__ constructor parses the large number of optional inputs into a model dictionary. In Consistency checks and mock-population bookkeeping we outline the various bookkeeping devices and consistency checks that the factory does in order to 1. ensure that the input model dictionary provides sufficient and self-consistent information, and 2. place the instance into a form that can directly talk to the SubhaloMockFactory. In Inheriting behaviors from the component models we cover the process by which the appropriate methods of the component models are inherited by the composite model. The syntax for using a composite model to create mock catalogs is covered in The populate_mock convenience method. We conclude in Further reading by pointing to sections of documentation covering related aspects such as the algorithm for using SubhaloModelFactory instances to populate mocks.

Overview of the factory design

The SubhaloModelFactory has virtually no behavior of its own; it should instead be thought of as a container class that collects together behaviors that are defined elsewhere. These behaviors are defined in component models, which are instances of Halotools classes that typically provide a single, specialized mapping between halos and some specific galaxy property. By composing these individual mappings together, the output of the factory is a composite model for the galaxy-halo connection in which any number of user-defined galaxy properties is simultaneously modeled.

Although there are numerous options for the form of the arguments passed to SubhaloModelFactory, the basic input is a model dictionary. A model dictionary is just an ordinary python dictionary that stores the collection of component model instances whose behaviors are being unified together by the factory. The model dictionary contains all of the necessary information inform the SubhaloModelFactory how to build a composite model from the components.

Each component model in a model dictionary typically has each of the following three private attributes:

  1. _methods_to_inherit

  2. _galprop_dtypes_to_allocate

  3. _mock_generation_calling_sequence

Each of these three attributes will be explained in detail below. Briefly, the _methods_to_inherit is a list of strings that instructs the SubhaloModelFactory which methods in the component model should be carried over into the composite model. The _galprop_dtypes_to_allocate attribute is used to instruct the SubhaloMockFactory of the shape and name of every Numpy array that should be allocated for every galaxy property assigned by the component model. The _mock_generation_calling_sequence specifies the sequential order in which the methods of the component model should be called by the composite model during mock population.

Again, we will discuss these and other bookkeeping devices in more detail below. For now, simply observe what is accomplished by these three pieces of information. Each component model is effectively giving the factory the following message: “I want you to know about the following methods, and only the following methods, and I will take care of how they will be computed: _methods_to_inherit; I need you to make sure that the when you call these methods, the following arrays that will be passed to them: _galprop_dtypes_to_allocate; when you use me to make a mock, I need you to call these methods in the following sequence: _mock_generation_calling_sequence”. In this way, not only is all the physically relevant behavior defined in the component models, but the component models themselves provide the instructions for how they should be used.

The job of the SubhaloModelFactory is simply to follow these instructions, and to ensure that mutually consistent messages are received from the set of components in the model dictionary. In the remaining sections of this tutorial, we will walk step-by-step through the tasks carried out when a new composite model is built by instantiating an instance of the SubhaloModelFactory class.

Inferring a model dictionary from the constructor inputs

The first thing the __init__ constructor of SubhaloModelFactory does is to pass all its arguments to the _parse_constructor_kwargs method, which simply extracts (if present) galaxy_selection_func, halo_selection_func and model_feature_calling_sequence from the arguments passed to __init__; all remaining arguments will be interpeted as model dictionary inputs. For an explanation of galaxy_selection_func and halo_selection_func, see the ModelFactory docstring.

When calling the constructor of the ModelFactory super-class after parsing the inputs, exact copies of all arguments passed to SubhaloModelFactory are bound to the instance. This allows all composite model instances to remember the exact set of instructions from which they were built. As we will see, this is useful because it simplifies the process of building alternate versions of any particular composite model instance.

As described in The model_feature_calling_sequence mechanism, the model_feature_calling_sequence determines the order in which the component models will be called during mock population. This order is determined by the build_model_feature_calling_sequence method.

Once this order is determined, the model_dictionary attribute is bound to the instance using the appropriate order:

self.model_dictionary = collections.OrderedDict()
for key in self._model_feature_calling_sequence:
    self.model_dictionary[key] = copy(self._input_model_dictionary[key])

In the next section, we will see how the model_dictionary attribute is used to create a number of bookkeeping mechanisms used to verify self-consistency between the model features, and also to facilitate communication between the composite model and the SubhaloMockFactory.

Consistency checks and mock-population bookkeeping

After the model dictionary has been built, the __init__ constructor creates a handful of lists and dictionaries and binds these to the instance with the following lines of code:

# Build up and bind several lists from the component models

These methods examine each of the component models, perform various self-consistency tests, and create standardized attributes that allow the composite model to communicate with the SubhaloMockFactory to populate mocks. For a description of the most important methods in this standardization process, see Composite Model Bookkeeping Mechanisms. At the end of this sequence of function calls, the instance is prepared to inherit the behavior of the primary methods of the component models, which we cover in the next section.

Inheriting behaviors from the component models

Once all of the above lists and dictionaries of the composite model have been created, the SubhaloModelFactory finally inherits the behaviors of the component models. This is done using with the set_primary_behaviors method.

This is the most important function in the entire factory. Although it is only a few lines, it is sufficiently complicated to warrant detailed discussion. First, we reproduce the source below:

for feature, component_model in self.model_dictionary.iteritems():

    for methodname in component_model._methods_to_inherit:

        new_method_name = methodname # line 1
        new_method_behavior = self.update_param_dict_decorator(
            component_model, methodname) # line 2
        setattr(self, new_method_name, new_method_behavior) # line 3

In this double-for loop, we iterate over every method that the composite model should inherit from the collection of component models. For each method that we inherit, line 3 binds the newly-defined method to the composite model instance. Line 1 chooses for the name of this newly-defined method to keep the same name as appears in the component model. Line 2 modifies the component model method behavior with the update_param_dict_decorator decorator. This modification is very important for the reasons described in The update_param_dict_decorator mechanism.

Note how the use of getattr and setattr allows the component models to entirely dictate what is inherited by the composite model. This high-level python feature is what makes possible the flexibility of the model factories.

The populate_mock convenience method

No matter what the component model features are, all instances of SubhaloModelFactory can directly populate subhalo catalogs with mock galaxies with the populate_mock method. To populate the default halo catalog, the syntax for this is:

model = SubhaloModelFactory(**model_dictionary)
from halotools.sim_manager import CachedHaloCatalog
halocat = CachedHaloCatalog()

The SubhaloModelFactory.populate_mock method is just a convenience wrapper around SubhaloMockFactory.populate method.

You can also populate alternative halo catalogs:

from halotools.sim_manager import CachedHaloCatalog
my_halocat = CachedHaloCatalog(simname = my_simname, redshift = my_redshift)

You can use the syntax above to populate any instance of either CachedHaloCatalog or UserSuppliedHaloCatalog.

Further reading

Detailed documentation on the mock-population algorithm is covered in Tutorial on the algorithm for subhalo-based mock-making.