__init__.py 11.1 KB
Newer Older
1
# -*- coding: utf-8 -*-
2
# Copyright (C) 2014 Laboratoire d'Astrophysique de Marseille, AMU
3
# Copyright (C) 2013 Centre de données Astrophysiques de Marseille
4 5
# Copyright (C) 2013-2014 Institute of Astronomy
# Copyright (C) 2013-2014 Yannick Roehlly <yannick@iaora.eu>
6
# Licensed under the CeCILL-v2 licence - see Licence_CeCILL_V2-en.txt
7
# Author: Yannick Roehlly, Médéric Boquien & Denis Burgarella
8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27

"""
Probability Density Function analysis module
============================================

This module builds the probability density functions (PDF) of the SED
parameters to compute their moments.

The models corresponding to all possible combinations of parameters are
computed and their fluxes in the same filters as the observations are
integrated. These fluxes are compared to the observed ones to compute the
χ² value of the fitting. This χ² give a probability that is associated with
the model values for the parameters.

At the end, for each parameter, the probability-weighted mean and standard
deviation are computed and the best fitting model (the one with the least
reduced χ²) is given for each observation.

"""

28
import ctypes
29
import multiprocessing as mp
30 31 32 33 34
from multiprocessing.sharedctypes import RawArray
import time

import numpy as np

35 36
from ...utils import read_table
from .. import AnalysisModule, complete_obs_table
37
from .utils import save_results, analyse_chi2
38
from ...warehouse import SedWarehouse
39
from .workers import sed as worker_sed
40 41
from .workers import init_sed as init_worker_sed
from .workers import init_analysis as init_worker_analysis
42
from .workers import analysis as worker_analysis
43
from ..utils import ParametersHandler, backup_dir
44

45

46
# Tolerance threshold under which any flux or error is considered as 0.
47
TOLERANCE = 1e-12
48 49 50 51 52


class PdfAnalysis(AnalysisModule):
    """PDF analysis module"""

53
    parameter_list = dict([
54 55 56 57
        ("analysed_variables", (
            "array of strings",
            "List of the variables (in the SEDs info dictionaries) for which "
            "the statistical analysis will be done.",
58
            ["sfh.sfr", "sfh.sfr10Myrs", "sfh.sfr100Myrs"]
59 60 61 62 63 64
        )),
        ("save_best_sed", (
            "boolean",
            "If true, save the best SED for each observation to a file.",
            False
        )),
65
        ("save_chi2", (
66
            "boolean",
67
            "If true, for each observation and each analysed variable save "
68
            "the reduced chi2.",
69 70 71 72
            False
        )),
        ("save_pdf", (
            "boolean",
73 74
            "If true, for each observation and each analysed variable save "
            "the probability density function.",
75 76
            False
        )),
77 78 79 80 81
        ("lim_flag", (
            "boolean",
            "If true, for each object check whether upper limits are present "
            "and analyse them.",
            False
82 83 84 85 86 87
        )),
        ("mock_flag", (
            "boolean",
            "If true, for each object we create a mock object "
            "and analyse them.",
            False
88 89 90
        ))
    ])

91
    def process(self, conf):
92 93
        """Process with the psum analysis.

94 95 96 97 98
        The analysis is done in two steps which can both run on multiple
        processors to run faster. The first step is to compute all the fluxes
        associated with each model as well as ancillary data such as the SED
        information. The second step is to carry out the analysis of each
        object, considering all models at once.
99 100 101

        Parameters
        ----------
102 103
        conf: dictionary
            Contents of pcigale.ini in the form of a dictionary
104 105

        """
106
        np.seterr(invalid='ignore')
107

108 109
        print("Initialising the analysis module... ")

110
        # Rename the output directory if it exists
111
        backup_dir()
112

113
        # Initalise variables from input arguments.
114 115 116
        creation_modules = conf['creation_modules']
        creation_modules_params = conf['creation_modules_params']
        analysed_variables = conf['analysis_method_params']["analysed_variables"]
117 118 119
        analysed_variables_nolog = [variable[:-4] if variable.endswith('_log')
                                    else variable for variable in
                                    analysed_variables]
120
        n_variables = len(analysed_variables)
121
        save = {key: conf['analysis_method_params']["save_{}".format(key)].lower() == "true"
122
                for key in ["best_sed", "chi2", "pdf"]}
123 124
        lim_flag = conf['analysis_method_params']["lim_flag"].lower() == "true"
        mock_flag = conf['analysis_method_params']["mock_flag"].lower() == "true"
125

126 127
        filters = [name for name in conf['column_list'] if not
                   name.endswith('_err')]
128
        n_filters = len(filters)
129 130 131

        # Read the observation table and complete it by adding error where
        # none is provided and by adding the systematic deviation.
132 133 134
        obs_table = complete_obs_table(read_table(conf['data_file']),
                                       conf['column_list'], filters, TOLERANCE,
                                       lim_flag)
135
        n_obs = len(obs_table)
136

137
        w_redshifting = creation_modules.index('redshifting')
138
        z = np.array(creation_modules_params[w_redshifting]['redshift'])
139 140 141 142 143 144 145 146 147

        # The parameters handler allows us to retrieve the models parameters
        # from a 1D index. This is useful in that we do not have to create
        # a list of parameters as they are computed on-the-fly. It also has
        # nice goodies such as finding the index of the first parameter to
        # have changed between two indices or the number of models.
        params = ParametersHandler(creation_modules, creation_modules_params)
        n_params = params.size

148
        # Retrieve an arbitrary SED to obtain the list of output parameters
149
        warehouse = SedWarehouse()
150
        sed = warehouse.get_sed(creation_modules, params.from_index(0))
151 152 153
        info = list(sed.info.keys())
        info.sort()
        n_info = len(info)
154 155
        del warehouse, sed

156 157
        print("Computing the models fluxes...")

158 159 160 161 162 163 164
        # Arrays where we store the data related to the models. For memory
        # efficiency reasons, we use RawArrays that will be passed in argument
        # to the pool. Each worker will fill a part of the RawArrays. It is
        # important that there is no conflict and that two different workers do
        # not write on the same section.
        # We put the shape in a tuple along with the RawArray because workers
        # need to know the shape to create the numpy array from the RawArray.
165
        model_fluxes = (RawArray(ctypes.c_double, n_params * n_filters),
166
                        (n_params, n_filters))
167
        model_variables = (RawArray(ctypes.c_double, n_params * n_variables),
168
                           (n_params, n_variables))
169

170 171
        initargs = (params, filters, analysed_variables_nolog, model_fluxes,
                    model_variables, time.time(), mp.Value('i', 0))
172
        if conf['cores'] == 1:  # Do not create a new process
173
            init_worker_sed(*initargs)
174 175
            for idx in range(n_params):
                worker_sed(idx)
176
        else:  # Compute the models in parallel
177
            with mp.Pool(processes=conf['cores'], initializer=init_worker_sed,
178
                         initargs=initargs) as pool:
179
                pool.map(worker_sed, range(n_params))
180

181
        print("\nAnalysing models...")
182

183 184
        # We use RawArrays for the same reason as previously
        analysed_averages = (RawArray(ctypes.c_double, n_obs * n_variables),
Médéric Boquien's avatar
Médéric Boquien committed
185
                             (n_obs, n_variables))
186
        analysed_std = (RawArray(ctypes.c_double, n_obs * n_variables),
Médéric Boquien's avatar
Médéric Boquien committed
187
                        (n_obs, n_variables))
188
        best_fluxes = (RawArray(ctypes.c_double, n_obs * n_filters),
Médéric Boquien's avatar
Médéric Boquien committed
189
                       (n_obs, n_filters))
190 191 192 193 194
        best_parameters = (RawArray(ctypes.c_double, n_obs * n_info),
                           (n_obs, n_info))
        best_chi2 = (RawArray(ctypes.c_double, n_obs), (n_obs))
        best_chi2_red = (RawArray(ctypes.c_double, n_obs), (n_obs))

195 196 197 198 199
        initargs = (params, filters, analysed_variables, z, model_fluxes,
                    model_variables, time.time(), mp.Value('i', 0),
                    analysed_averages, analysed_std, best_fluxes,
                    best_parameters, best_chi2, best_chi2_red, save, lim_flag,
                    n_obs)
200
        if conf['cores'] == 1:  # Do not create a new process
201
            init_worker_analysis(*initargs)
202 203
            for idx, obs in enumerate(obs_table):
                worker_analysis(idx, obs)
204
        else:  # Analyse observations in parallel
205 206
            with mp.Pool(processes=conf['cores'],
                         initializer=init_worker_analysis,
207
                         initargs=initargs) as pool:
208
                pool.starmap(worker_analysis, enumerate(obs_table))
209

210
        analyse_chi2(best_chi2_red)
211

212 213
        print("\nSaving results...")

214 215 216
        save_results("results", obs_table['id'], analysed_variables,
                     analysed_averages, analysed_std, best_chi2, best_chi2_red,
                     best_parameters, best_fluxes, filters, info)
217 218 219 220 221

        if mock_flag is True:

            print("\nMock analysis...")

222 223 224 225
            # For the mock analysis we do not save the ancillary files
            for k in save:
                save[k] = False

226
            obs_fluxes = np.array([obs_table[name] for name in filters]).T
Médéric Boquien's avatar
Médéric Boquien committed
227 228
            obs_errors = np.array([obs_table[name + "_err"] for name in
                                   filters]).T
229
            mock_fluxes = obs_fluxes.copy()
230 231
            bestmod_fluxes = np.ctypeslib.as_array(best_fluxes[0])
            bestmod_fluxes = bestmod_fluxes.reshape(best_fluxes[1])
232 233 234 235
            wdata = np.where((obs_fluxes > TOLERANCE) &
                             (obs_errors > TOLERANCE))
            mock_fluxes[wdata] = np.random.normal(bestmod_fluxes[wdata],
                                                  obs_errors[wdata])
Médéric Boquien's avatar
Médéric Boquien committed
236

237
            mock_table = obs_table.copy()
238 239
            for idx, name in enumerate(filters):
                mock_table[name] = mock_fluxes[:, idx]
Médéric Boquien's avatar
Médéric Boquien committed
240

241 242 243 244 245
            initargs = (params, filters, analysed_variables, z, model_fluxes,
                        model_variables, time.time(), mp.Value('i', 0),
                        analysed_averages, analysed_std, best_fluxes,
                        best_parameters, best_chi2, best_chi2_red, save,
                        lim_flag, n_obs)
246 247 248 249 250 251 252 253 254
            if cores == 1:  # Do not create a new process
                init_worker_analysis(*initargs)
                for idx, mock in enumerate(mock_table):
                    worker_analysis(idx, mock)
            else:  # Analyse observations in parallel
                with mp.Pool(processes=cores, initializer=init_worker_analysis,
                             initargs=initargs) as pool:
                    pool.starmap(worker_analysis, enumerate(mock_table))

255
            print("\nSaving results...")
256

257 258 259 260
            save_results("results_mock", mock_table['id'], analysed_variables,
                         analysed_averages, analysed_std, best_chi2,
                         best_chi2_red, best_parameters, best_fluxes, filters,
                         info)
261 262

        print("Run completed!")
263

264 265
# AnalysisModule to be returned by get_module
Module = PdfAnalysis