Source code for halotools.empirical_models.factories.hod_mock_factory

"""
Module containing the `~halotools.empirical_models.HodMockFactory` class,
the primary class used to construct mock galaxy populations
based on HOD-style models.

The `~halotools.empirical_models.HodMockFactory` class
provides an abstract interface between halo catalogs
and Halotools models.
"""

import numpy as np
from copy import copy
from astropy.table import Table
from astropy.utils.misc import NumpyRNGContext

from .mock_factory_template import MockFactory

from .. import model_helpers

from ...sim_manager import sim_defaults
from ...utils.table_utils import SampleSelector
from ...custom_exceptions import HalotoolsError


__all__ = ['HodMockFactory']
__author__ = ['Andrew Hearin']

unavailable_haloprop_msg = ("Your model requires that the ``%s`` key appear in the halo catalog,\n"
    "but this column is not available in the catalog you attempted to populate.\n")

missing_halo_upid_msg = ("All HOD-style models populate host halos with mock galaxies.\n"
    "The way Halotools distinguishes host halos from subhalos is via the ``halo_upid`` column,\n"
    "with halo_upid = -1 for host halos and !=-1 for subhalos.\n"
    "The halo catalog you passed to the HodMockFactory does not have the ``halo_upid`` column.\n")


[docs]class HodMockFactory(MockFactory): """ Class responsible for populating a simulation with a population of mock galaxies based on an HOD-style model built by the `~halotools.empirical_models.HodModelFactory` class. Can be thought of as a factory that takes a model and simulation halocat as input, and generates a mock galaxy population. The returned collection of galaxies possesses whatever attributes were requested by the model, such as xyz position, central/satellite designation, star-formation rate, etc. See :ref:`hod_mock_factory_source_notes` for an in-depth tutorial on the mock-making algorithm. """ def __init__(self, Num_ptcl_requirement=sim_defaults.Num_ptcl_requirement, halo_mass_column_key='halo_mvir', **kwargs): """ Parameters ---------- halocat : object, keyword argument Object containing the halo catalog and other associated data. Produced by `~halotools.sim_manager.CachedHaloCatalog` model : object, keyword argument A model built by a sub-class of `~halotools.empirical_models.HodModelFactory`. populate : boolean, optional If set to ``False``, the class will perform all pre-processing tasks but will not call the ``model`` to populate the ``galaxy_table`` with mock galaxies and their observable properties. Default is ``True``. Num_ptcl_requirement : int, optional Requirement on the number of dark matter particles in the halo. The column defined by the ``halo_mass_column_key`` string will have a cut placed on it: all halos with halocat.halo_table[halo_mass_column_key] < Num_ptcl_requirement*halocat.particle_mass will be thrown out immediately after reading the original halo catalog in memory. Default value is set in `~halotools.sim_defaults.Num_ptcl_requirement`. halo_mass_column_key : string, optional This string must be a column of the input halo catalog. The column defined by this string will have a cut placed on it: all halos with halocat.halo_table[halo_mass_column_key] < Num_ptcl_requirement*halocat.particle_mass will be thrown out immediately after reading the original halo catalog in memory. Default is 'halo_mvir' """ MockFactory.__init__(self, **kwargs) halocat = kwargs['halocat'] self.Num_ptcl_requirement = Num_ptcl_requirement self.halo_mass_column_key = halo_mass_column_key self.preprocess_halo_catalog(halocat)
[docs] def preprocess_halo_catalog(self, halocat): """ Method to pre-process a halo catalog upon instantiation of the mock object. This pre-processing includes identifying the catalog columns that will be used by the model to create the mock, building lookup tables associated with the halo profile, and possibly creating new halo properties. Parameters ---------- logrmin : float, optional Minimum radius used to build the lookup table for the halo profile. Default is set in `~halotools.empirical_models.model_defaults`. logrmax : float, optional Maximum radius used to build the lookup table for the halo profile. Default is set in `~halotools.empirical_models.model_defaults`. Npts_radius_table : int, optional Number of control points used in the lookup table for the halo profile. Default is set in `~halotools.empirical_models.model_defaults`. """ try: assert 'halo_upid' in list(halocat.halo_table.keys()) except AssertionError: raise HalotoolsError(missing_halo_upid_msg) # Make cuts on halo catalog # # Select host halos only, since this is an HOD-style model halo_table, subhalo_table = SampleSelector.host_halo_selection( table=halocat.halo_table, return_subhalos=True) # make a (possibly trivial) completeness cut cutoff_mvir = self.Num_ptcl_requirement*self.particle_mass mass_cut = halo_table[self.halo_mass_column_key] > cutoff_mvir max_column_value = np.max(halo_table[self.halo_mass_column_key]) halo_table = halo_table[mass_cut] if len(halo_table) == 0: msg = ("During the pre-processing phase of HOD mock-making \n" "controlled by the `preprocess_halo_catalog` method of the HodMockFactory,\n" "a cut on halo completeness is made. The column used in this cut\n" "is determined by the value of the halo_mass_column_key string \n" "passed to the HodMockFactory constructor, which in this case was ``%s``.\n" "The largest value of this column in the uncut catalog is %.2e.\n" "Your mass cut of M > %.2e resulted in zero halos in the halo catalog;\n" "for reference, your halo catalog has a particle mass of m_p = %.2e.\n" "Such a cut is not permissible. \nThis could indicate a problem in " "the processing of the halo catalog,\n" "for example, an incorrect column number and/or dtype.\n" "Alternatively, the value of Num_ptcl_requirement = %.2e \n" "that was passed to the HodMockFactory constructor could be the problem.\n") raise HalotoolsError(msg % ( self.halo_mass_column_key, max_column_value, cutoff_mvir, self.particle_mass, self.Num_ptcl_requirement)) # If any component model needs the subhalo_table, bind it to the mock object for component_model in list(self.model.model_dictionary.values()): try: f = getattr(component_model, 'preprocess_subhalo_table') halo_table, self.subhalo_table = f(halo_table, subhalo_table) except AttributeError: pass ############################################################ # Create new columns of the halo catalog, if applicable try: d = self.model.new_haloprop_func_dict for new_haloprop_key, new_haloprop_func in list(d.items()): halo_table[new_haloprop_key] = new_haloprop_func(table=halo_table) self.additional_haloprops.append(new_haloprop_key) except AttributeError: pass self._orig_halo_table = Table() for key in self.additional_haloprops: try: self._orig_halo_table[key] = halo_table[key][:] except KeyError: raise HalotoolsError(unavailable_haloprop_msg % key) self.model.build_lookup_tables()
[docs] def populate(self, seed=None, **kwargs): """ Method populating host halos with mock galaxies. By calling the `populate` method of your mock, you will repopulate the halo catalog with a new realization of the model based on whatever values of the model parameters are currently stored in the ``param_dict`` of the model. For an in-depth discussion of how this method is implemented, see the :ref:`hod_mock_factory_source_notes` section of the documentation. Parameters ------------ masking_function : function, optional Function object used to place a mask on the halo table prior to calling the mock generating functions. Calling signature of the function should be to accept a single positional argument storing a table, and returning a boolean numpy array that will be used as a fancy indexing mask. All masked halos will be ignored during mock population. Default is None. enforce_PBC : bool, optional If set to True, after galaxy positions are assigned the `model_helpers.enforce_periodicity_of_box` will re-map satellite galaxies whose positions spilled over the edge of the periodic box. Default is True. This variable should only ever be set to False when using the ``masking_function`` to populate a specific spatial subvolume, as in that case PBCs no longer apply. seed : int, optional Random number seed used in the Monte Carlo realization. Default is None, which will produce stochastic results. Notes ----- Note the difference between the `halotools.empirical_models.HodMockFactory.populate` method and the closely related method `halotools.empirical_models.HodModelFactory.populate_mock`. The `~halotools.empirical_models.HodModelFactory.populate_mock` method is bound to a composite model instance and is called the *first* time a composite model is used to generate a mock. Calling the `~halotools.empirical_models.HodModelFactory.populate_mock` method creates the `~halotools.empirical_models.HodMockFactory` instance and binds it to composite model. From then on, if you want to *repopulate* a new Universe with the same composite model, you should instead call the `~halotools.empirical_models.HodMockFactory.populate` method bound to ``model.mock``. The reason for this distinction is that calling `~halotools.empirical_models.HodModelFactory.populate_mock` triggers a large number of relatively expensive pre-processing steps and self-consistency checks that need only be carried out once. See the Examples section below for an explicit demonstration. In particular, if you are running an MCMC type analysis, you will choose your halo catalog and completeness cuts, and call `halotools.empirical_models.ModelFactory.populate_mock` with the appropriate arguments. Thereafter, you can explore parameter space by changing the values stored in the ``param_dict`` dictionary attached to the model, and then calling the `~halotools.empirical_models.MockFactory.populate` method bound to ``model.mock``. Any changes to the ``param_dict`` of the model will automatically propagate into the behavior of the `~halotools.empirical_models.MockFactory.populate` method. Normally, repeated calls to the `~halotools.empirical_models.HodMockFactory.populate` method should not increase the RAM usage of halotools because a new mock catalog is created and the old one deleted. However, on certain machines the memory usage was found to increase over time. If this is the case and memory usage is critical you can try calling gc.collect() immediately following the call to ``mock.populate`` to manually invoke python's garbage collection. Examples ---------- >>> from halotools.empirical_models import PrebuiltHodModelFactory >>> model_instance = PrebuiltHodModelFactory('zheng07') Here we will use a fake simulation, but you can populate mocks using any instance of `~halotools.sim_manager.CachedHaloCatalog` or `~halotools.sim_manager.UserSuppliedHaloCatalog`. >>> from halotools.sim_manager import FakeSim >>> halocat = FakeSim() >>> model_instance.populate_mock(halocat) Your ``model_instance`` now has a ``mock`` attribute bound to it, which is an instance of the `~halotools.empirical_models.HodMockFactory` class. You can call the `populate` method bound to the ``mock``, which will repopulate the halo catalog with a new Monte Carlo realization of the model. >>> model_instance.mock.populate() If you want to change the behavior of your model, just change the values stored in the ``param_dict``. The ``param_dict`` attribute is a python dictionary storing the values of all parameters in the model. Differences in the parameter values will change the behavior of the mock-population. >>> model_instance.param_dict['logMmin'] = 12.1 >>> model_instance.mock.populate() See also --------- :ref:`hod_mock_factory_source_notes` """ # The _testing_mode keyword is for unit-testing only # it has been intentionally left out of the docstring try: self._testing_mode = kwargs['_testing_mode'] except KeyError: self._testing_mode = False try: self.enforce_PBC = kwargs['enforce_PBC'] except KeyError: self.enforce_PBC = True try: masking_function = kwargs['masking_function'] mask = masking_function(self._orig_halo_table) self.halo_table = self._orig_halo_table[mask] except: self.halo_table = self._orig_halo_table self.allocate_memory(seed=seed) # Loop over all gal_types in the model for gal_type in self.gal_types: # Retrieve the indices of our pre-allocated arrays # that store the info pertaining to gal_type galaxies gal_type_slice = self._gal_type_indices[gal_type] # gal_type_slice is a slice object # For the gal_type_slice indices of # the pre-allocated array self.gal_type, # set each string-type entry equal to the gal_type string self.galaxy_table['gal_type'][gal_type_slice] = ( np.repeat(gal_type, self._total_abundance[gal_type], axis=0)) # Store all other relevant host halo properties into their # appropriate pre-allocated array for halocatkey in self.additional_haloprops: self.galaxy_table[halocatkey][gal_type_slice] = np.repeat( self.halo_table[halocatkey], self._occupation[gal_type], axis=0) self.galaxy_table['x'] = self.galaxy_table['halo_x'] self.galaxy_table['y'] = self.galaxy_table['halo_y'] self.galaxy_table['z'] = self.galaxy_table['halo_z'] self.galaxy_table['vx'] = self.galaxy_table['halo_vx'] self.galaxy_table['vy'] = self.galaxy_table['halo_vy'] self.galaxy_table['vz'] = self.galaxy_table['halo_vz'] for method in self._remaining_methods_to_call: func = getattr(self.model, method) try: d = {key: getattr(self, key) for key in func.additional_kwargs} except AttributeError: d = {} gal_type_slice = self._gal_type_indices[func.gal_type] if seed is not None: seed += 1 func(table=self.galaxy_table[gal_type_slice], seed=seed, **d) if self.enforce_PBC is True: self.galaxy_table['x'], self.galaxy_table['vx'] = ( model_helpers.enforce_periodicity_of_box( self.galaxy_table['x'], self.Lbox[0], velocity=self.galaxy_table['vx'], check_multiple_box_lengths=self._testing_mode) ) self.galaxy_table['y'], self.galaxy_table['vy'] = ( model_helpers.enforce_periodicity_of_box( self.galaxy_table['y'], self.Lbox[1], velocity=self.galaxy_table['vy'], check_multiple_box_lengths=self._testing_mode) ) self.galaxy_table['z'], self.galaxy_table['vz'] = ( model_helpers.enforce_periodicity_of_box( self.galaxy_table['z'], self.Lbox[2], velocity=self.galaxy_table['vz'], check_multiple_box_lengths=self._testing_mode) ) if hasattr(self.model, 'galaxy_selection_func'): mask = self.model.galaxy_selection_func(self.galaxy_table) self.galaxy_table = self.galaxy_table[mask]
[docs] def allocate_memory(self, seed=None): """ Method allocates the memory for all the numpy arrays that will store the information about the mock. These arrays are bound directly to the mock object. The main bookkeeping devices generated by this method are ``_occupation`` and ``_gal_type_indices``. """ self.galaxy_table = Table() # We will keep track of the calling sequence with a list called _remaining_methods_to_call # Each time a function in this list is called, we will remove that function from the list # Mock generation will be complete when _remaining_methods_to_call is exhausted self._remaining_methods_to_call = copy(self.model._mock_generation_calling_sequence) # Call all composite model methods that should be called prior to mc_occupation # All such function calls must be applied to the table, since we do not yet know # how much memory we need for the mock galaxy_table galprops_assigned_to_halo_table = [] for func_name in self.model._mock_generation_calling_sequence: if 'mc_occupation' in func_name: # exit when we encounter a ``mc_occupation_`` function break else: func = getattr(self.model, func_name) try: d = {key: getattr(self, key) for key in func.additional_kwargs} except AttributeError: d = {} if seed is not None: seed += 1 func(table=self.halo_table, seed=seed, **d) galprops_assigned_to_halo_table_by_func = func._galprop_dtypes_to_allocate.names galprops_assigned_to_halo_table.extend(galprops_assigned_to_halo_table_by_func) self._remaining_methods_to_call.remove(func_name) # Now update the list of additional_haloprops, if applicable # This is necessary because each of the above function calls created new # columns for the *halo_table*, not the *galaxy_table*. So we will need to use # np.repeat inside mock.populate() so that mock galaxies inherit these newly-created columns # Since there is already a loop over additional_haloprops inside mock.populate() that does this, # then all we need to do is append to this list galprops_assigned_to_halo_table = list(set( galprops_assigned_to_halo_table)) self.additional_haloprops.extend(galprops_assigned_to_halo_table) self.additional_haloprops = list(set(self.additional_haloprops)) self._occupation = {} self._total_abundance = {} self._gal_type_indices = {} for gal_type in self.gal_types: self.halo_table['halo_num_'+gal_type] = 0 first_galaxy_index = 0 for gal_type in self.gal_types: occupation_func_name = 'mc_occupation_'+gal_type occupation_func = getattr(self.model, occupation_func_name) # Call the component model to get a Monte Carlo # realization of the abundance of gal_type galaxies if seed is not None: seed += 1 self._occupation[gal_type] = occupation_func(table=self.halo_table, seed=seed) self.halo_table['halo_num_'+gal_type][:] = self._occupation[gal_type] # Now use the above result to set up the indexing scheme self._total_abundance[gal_type] = ( self._occupation[gal_type].sum() ) last_galaxy_index = first_galaxy_index + self._total_abundance[gal_type] # Build a bookkeeping device to keep track of # which array elements pertain to which gal_type. self._gal_type_indices[gal_type] = slice( first_galaxy_index, last_galaxy_index) first_galaxy_index = last_galaxy_index # Remove the mc_occupation function from the list of methods to call self._remaining_methods_to_call.remove(occupation_func_name) self.additional_haloprops.append('halo_num_'+gal_type) self.Ngals = np.sum(list(self._total_abundance.values())) # Allocate memory for all additional halo properties, # including profile parameters of the halos such as 'conc_NFWmodel' for halocatkey in self.additional_haloprops: self.galaxy_table[halocatkey] = np.zeros(self.Ngals, dtype=self.halo_table[halocatkey].dtype) # Separately allocate memory for the galaxy profile parameters for galcatkey in self.model.halo_prof_param_keys: self.galaxy_table[galcatkey] = 0. for galcatkey in self.model.gal_prof_param_keys: self.galaxy_table[galcatkey] = 0. self.galaxy_table['gal_type'] = np.zeros(self.Ngals, dtype=object) dt = self.model._galprop_dtypes_to_allocate for key in dt.names: self.galaxy_table[key] = np.zeros(self.Ngals, dtype=dt[key].type)
[docs] def estimate_ngals(self, seed=None): """ Method to estimate the number of galaxies produced by the mock.populate() method. It runs one realization of all mc_occupation methods and reports the total number of galaxies produced. However, no extra memory is allocated for the galaxy tables. Note that model.populate() will invoke a new call to all mc_occupation methods and can produce a different number of galaxies. """ # Call all composite model methods that should be called prior to mc_occupation # All such function calls must be applied to the table. halo_table = Table(np.copy(self.halo_table)) for func_name in self.model._mock_generation_calling_sequence: if 'mc_occupation' in func_name: # exit when we encounter a ``mc_occupation_`` function break else: func = getattr(self.model, func_name) func(table=halo_table) # Call the component model to get a Monte Carlo # realization of the abundance of galaxies for all gal_type. ngals = 0 for gal_type in self.gal_types: occupation_func_name = 'mc_occupation_'+gal_type occupation_func = getattr(self.model, occupation_func_name) if seed is not None: seed += 1 ngals = ngals + np.sum(occupation_func(table=halo_table, seed=seed)) return ngals