# -*- coding: utf-8 -*-
"""
Define the sampler class which reads goes from the distribution dictionary to
drawing samples on the lens parameters.
This module contains the class used to sample parameters for our train and test
set from the input distributions.
"""
import warnings
# Definte the components we need the sampler to consider.
# TODO: add point source & lens light
lensing_components = ['subhalo','los','main_deflector','source','lens_light',
'point_source','lens_equation_solver','cosmology','psf','detector','drizzle']
# Global filters on the python warnings. Using this since filter
# behaviour is a bit weird.
CROSSOBJECTWARNING = True
[docs]
class Sampler():
"""Class for drawing lens parameter values from input distribution
dictionaries
Args:
configuration_dictionary (dict): An instance of the configuration
dictionary that will be used to decide how to sample parameter
values.
Notes:
For examples on how to configure the dict object, see the
ConfigurationDict.ipynb example notebook.
"""
def __init__(self,configuration_dictionary):
self.config_dict = configuration_dictionary
[docs]
@staticmethod
def draw_from_dict(draw_dict):
"""Populates a dict with samples drawn from the specified distributions
in the input dict.
Args:
draw_dict (dict): The dictionary containing keys mapping to values
or distributions for each parameter.
Returns:
(dict): A dict with a drawn value for each parameter.
Notes:
Multivariate distribution for parameters should have a key of the
form 'param_1,param_2,param_3'.
"""
param_dict = {}
# Iterate through the keys in the draw_dict and populate the values of
# param_dict correctly.
for key in sorted(draw_dict):
# If the key implies that multiple parameters will be drawn from
# the distribution, draw the value and then iterate through the
# parameters.
if ',' in key:
# Get the parameters, removing whitespace
params = key.replace(' ','').split(',')
# Draw the values
draw = draw_dict[key]()
# Check for consistency
if len(params) != len(draw):
raise ValueError('Parameters of length %d do'%(len(params))
+ ' not match draw of length %d'%(len(draw)))
# Populate the keys
for i, param in enumerate(params):
param_dict[param] = draw[i]
# If it's a univariate function just call it.
elif callable(draw_dict[key]):
param_dict[key] = draw_dict[key]()
# If it's a fixed value just populate it.
else:
param_dict[key] = draw_dict[key]
return param_dict
[docs]
def sample(self):
"""Samples from the distributions given in the configuration
dictionary
Returns:
(dict): A dictionary containing the parameter values that will
be sampled.
"""
# Pull the global warning variable and initialize our dict
global CROSSOBJECTWARNING
full_param_dict = {}
# For each possible component of our lensing add the parameters
for component in lensing_components:
if component in self.config_dict:
draw_dict = self.config_dict[component]['parameters']
param_dict = self.draw_from_dict(draw_dict)
full_param_dict[component+'_parameters'] = param_dict
# Populate parameters from distributions that span accross objects.
if 'cross_object' in self.config_dict:
draw_dict = self.config_dict['cross_object']['parameters']
cross_dict = self.draw_from_dict(draw_dict)
# Go through the params and update the full dict
for cross_param in cross_dict:
component, param = cross_param.split(':')
param_dict = full_param_dict[component+'_parameters']
# Warn the user the first time an overwrite happens
if param in param_dict and CROSSOBJECTWARNING:
warnings.warn('Parameter %s in cross dict specified '%(param)
+ 'elsewhere! Will be overwritten. This warning only ' +
'flags once, but other parameters may also be ' +
'overwritten.')
CROSSOBJECTWARNING = False
full_param_dict[component+'_parameters'][param] = (
cross_dict[cross_param])
# Populate the cross objects
return full_param_dict