result.py 42.4 KB
Newer Older
Colm Talbot's avatar
Colm Talbot committed
1 2
from __future__ import division

3
import os
4
from distutils.version import LooseVersion
5
from collections import OrderedDict, namedtuple
MoritzThomasHuebner's avatar
MoritzThomasHuebner committed
6

7
import numpy as np
8
import deepdish
9
import pandas as pd
10
import corner
11
import scipy.stats
12
import matplotlib
13
import matplotlib.pyplot as plt
14

15
from . import utils
Colm Talbot's avatar
Colm Talbot committed
16 17 18
from .utils import (logger, infer_parameters_from_function,
                    check_directory_exists_and_if_not_mkdir)
from .prior import Prior, PriorDict, DeltaFunction
19

20

21
def result_file_name(outdir, label):
22 23 24 25 26 27 28 29 30 31 32 33 34
    """ Returns the standard filename used for a result file

    Parameters
    ----------
    outdir: str
        Name of the output directory
    label: str
        Naming scheme of the output file

    Returns
    -------
    str: File name of the output file
    """
35 36 37
    return '{}/{}_result.h5'.format(outdir, label)


38
def read_in_result(filename=None, outdir=None, label=None):
Gregory Ashton's avatar
Gregory Ashton committed
39 40 41 42 43 44
    """ Read in a saved .h5 data file

    Parameters
    ----------
    filename: str
        If given, try to load from this filename
45 46
    outdir, label: str
        If given, use the default naming convention for saved results file
Gregory Ashton's avatar
Gregory Ashton committed
47

48 49
    Returns
    -------
Colm Talbot's avatar
Colm Talbot committed
50
    result: bilby.core.result.Result
51

Moritz Huebner's avatar
Moritz Huebner committed
52
    Raises
53 54
    -------
    ValueError: If no filename is given and either outdir or label is None
Colm Talbot's avatar
Colm Talbot committed
55
                If no bilby.core.result.Result is found in the path
Gregory Ashton's avatar
Gregory Ashton committed
56 57 58

    """
    if filename is None:
59 60 61 62
        if (outdir is None) and (label is None):
            raise ValueError("No information given to load file")
        else:
            filename = result_file_name(outdir, label)
63
    if os.path.isfile(filename):
64
        return Result(**deepdish.io.load(filename))
65
    else:
66 67 68 69 70 71 72 73 74 75 76
        raise IOError("No result '{}' found".format(filename))


class Result(object):
    def __init__(self, label='no_label', outdir='.', sampler=None,
                 search_parameter_keys=None, fixed_parameter_keys=None,
                 priors=None, sampler_kwargs=None, injection_parameters=None,
                 meta_data=None, posterior=None, samples=None,
                 nested_samples=None, log_evidence=np.nan,
                 log_evidence_err=np.nan, log_noise_evidence=np.nan,
                 log_bayes_factor=np.nan, log_likelihood_evaluations=None,
77 78 79 80
                 log_prior_evaluations=None, sampling_time=None, nburn=None,
                 walkers=None, max_autocorrelation_time=None,
                 parameter_labels=None, parameter_labels_with_unit=None,
                 version=None):
81
        """ A class to store the results of the sampling run
82 83 84

        Parameters
        ----------
85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105
        label, outdir, sampler: str
            The label, output directory, and sampler used
        search_parameter_keys, fixed_parameter_keys: list
            Lists of the search and fixed parameter keys. Elemenents of the
            list should be of type `str` and matchs the keys of the `prior`
        priors: dict, bilby.core.prior.PriorDict
            A dictionary of the priors used in the run
        sampler_kwargs: dict
            Key word arguments passed to the sampler
        injection_parameters: dict
            A dictionary of the injection parameters
        meta_data: dict
            A dictionary of meta data to store about the run
        posterior: pandas.DataFrame
            A pandas data frame of the posterior
        samples, nested_samples: array_like
            An array of the output posterior samples and the unweighted samples
        log_evidence, log_evidence_err, log_noise_evidence, log_bayes_factor: float
            Natural log evidences
        log_likelihood_evaluations: array_like
            The evaluations of the likelihood for each sample point
106 107
        log_prior_evaluations: array_like
            The evaluations of the prior for each sample point
108 109 110 111 112 113 114 115 116 117
        sampling_time: float
            The time taken to complete the sampling
        nburn: int
            The number of burn-in steps discarded for MCMC samplers
        walkers: array_like
            The samplers taken by a ensemble MCMC samplers
        max_autocorrelation_time: float
            The estimated maximum autocorrelation time for MCMC samplers
        parameter_labels, parameter_labels_with_unit: list
            Lists of the latex-formatted parameter labels
118 119 120
        version: str,
            Version information for software used to generate the result. Note,
            this information is generated when the result object is initialized
121 122 123 124

        Note:
            All sampling output parameters, e.g. the samples themselves are
            typically not given at initialisation, but set at a later stage.
125 126

        """
127

128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148
        self.label = label
        self.outdir = os.path.abspath(outdir)
        self.sampler = sampler
        self.search_parameter_keys = search_parameter_keys
        self.fixed_parameter_keys = fixed_parameter_keys
        self.parameter_labels = parameter_labels
        self.parameter_labels_with_unit = parameter_labels_with_unit
        self.priors = priors
        self.sampler_kwargs = sampler_kwargs
        self.meta_data = meta_data
        self.injection_parameters = injection_parameters
        self.posterior = posterior
        self.samples = samples
        self.nested_samples = nested_samples
        self.walkers = walkers
        self.nburn = nburn
        self.log_evidence = log_evidence
        self.log_evidence_err = log_evidence_err
        self.log_noise_evidence = log_noise_evidence
        self.log_bayes_factor = log_bayes_factor
        self.log_likelihood_evaluations = log_likelihood_evaluations
149
        self.log_prior_evaluations = log_prior_evaluations
150
        self.sampling_time = sampling_time
151
        self.version = version
Colm Talbot's avatar
Colm Talbot committed
152
        self.max_autocorrelation_time = max_autocorrelation_time
153 154

    def __str__(self):
155
        """Print a summary """
156 157
        if getattr(self, 'posterior', None) is not None:
            if getattr(self, 'log_noise_evidence', None) is not None:
158 159 160 161
                return ("nsamples: {:d}\n"
                        "log_noise_evidence: {:6.3f}\n"
                        "log_evidence: {:6.3f} +/- {:6.3f}\n"
                        "log_bayes_factor: {:6.3f} +/- {:6.3f}\n"
162
                        .format(len(self.posterior), self.log_noise_evidence, self.log_evidence,
163 164 165 166 167
                                self.log_evidence_err, self.log_bayes_factor,
                                self.log_evidence_err))
            else:
                return ("nsamples: {:d}\n"
                        "log_evidence: {:6.3f} +/- {:6.3f}\n"
168
                        .format(len(self.posterior), self.log_evidence, self.log_evidence_err))
169 170
        else:
            return ''
171

172 173 174 175 176 177
    @property
    def priors(self):
        if self._priors is not None:
            return self._priors
        else:
            raise ValueError('Result object has no priors')
178

179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196
    @priors.setter
    def priors(self, priors):
        if isinstance(priors, dict):
            self._priors = PriorDict(priors)
            if self.parameter_labels is None:
                self.parameter_labels = [self.priors[k].latex_label for k in
                                         self.search_parameter_keys]
            if self.parameter_labels_with_unit is None:
                self.parameter_labels_with_unit = [
                    self.priors[k].latex_label_with_unit for k in
                    self.search_parameter_keys]

        elif priors is None:
            self._priors = priors
            self.parameter_labels = self.search_parameter_keys
            self.parameter_labels_with_unit = self.search_parameter_keys
        else:
            raise ValueError("Input priors not understood")
197

198 199 200 201 202
    @property
    def samples(self):
        """ An array of samples """
        if self._samples is not None:
            return self._samples
203
        else:
204
            raise ValueError("Result object has no stored samples")
205

206 207 208
    @samples.setter
    def samples(self, samples):
        self._samples = samples
209

210 211 212 213 214 215 216
    @property
    def nested_samples(self):
        """" An array of unweighted samples """
        if self._nested_samples is not None:
            return self._nested_samples
        else:
            raise ValueError("Result object has no stored nested samples")
217

218 219 220
    @nested_samples.setter
    def nested_samples(self, nested_samples):
        self._nested_samples = nested_samples
221

222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257
    @property
    def walkers(self):
        """" An array of the ensemble walkers """
        if self._walkers is not None:
            return self._walkers
        else:
            raise ValueError("Result object has no stored walkers")

    @walkers.setter
    def walkers(self, walkers):
        self._walkers = walkers

    @property
    def nburn(self):
        """" An array of the ensemble walkers """
        if self._nburn is not None:
            return self._nburn
        else:
            raise ValueError("Result object has no stored nburn")

    @nburn.setter
    def nburn(self, nburn):
        self._nburn = nburn

    @property
    def posterior(self):
        """ A pandas data frame of the posterior """
        if self._posterior is not None:
            return self._posterior
        else:
            raise ValueError("Result object has no stored posterior")

    @posterior.setter
    def posterior(self, posterior):
        self._posterior = posterior

258 259 260 261 262 263 264 265 266 267 268
    @property
    def version(self):
        return self._version

    @version.setter
    def version(self, version):
        if version is None:
            self._version = 'bilby={}'.format(utils.get_version_information())
        else:
            self._version = version

269
    def _get_save_data_dictionary(self):
270
        # This list defines all the parameters saved in the result object
271 272 273 274 275
        save_attrs = [
            'label', 'outdir', 'sampler', 'log_evidence', 'log_evidence_err',
            'log_noise_evidence', 'log_bayes_factor', 'priors', 'posterior',
            'injection_parameters', 'meta_data', 'search_parameter_keys',
            'fixed_parameter_keys', 'sampling_time', 'sampler_kwargs',
276 277
            'log_likelihood_evaluations', 'log_prior_evaluations', 'samples',
            'nested_samples', 'walkers', 'nburn', 'parameter_labels',
278
            'parameter_labels_with_unit', 'version']
279 280 281 282 283 284 285 286
        dictionary = OrderedDict()
        for attr in save_attrs:
            try:
                dictionary[attr] = getattr(self, attr)
            except ValueError as e:
                logger.debug("Unable to save {}, message: {}".format(attr, e))
                pass
        return dictionary
287

Colm Talbot's avatar
Colm Talbot committed
288 289 290 291 292 293 294 295 296 297
    def save_to_file(self, overwrite=False):
        """
        Writes the Result to a deepdish h5 file

        Parameters
        ----------
        overwrite: bool, optional
            Whether or not to overwrite an existing result file.
            default=False
        """
298
        file_name = result_file_name(self.outdir, self.label)
299
        utils.check_directory_exists_and_if_not_mkdir(self.outdir)
300
        if os.path.isfile(file_name):
Colm Talbot's avatar
Colm Talbot committed
301 302 303 304 305 306 307 308
            if overwrite:
                logger.debug('Removing existing file {}'.format(file_name))
                os.remove(file_name)
            else:
                logger.debug(
                    'Renaming existing file {} to {}.old'.format(file_name,
                                                                 file_name))
                os.rename(file_name, file_name + '.old')
309

Gregory Ashton's avatar
Gregory Ashton committed
310
        logger.debug("Saving result to {}".format(file_name))
311 312

        # Convert the prior to a string representation for saving on disk
313
        dictionary = self._get_save_data_dictionary()
314 315 316
        if dictionary.get('priors', False):
            dictionary['priors'] = {key: str(self.priors[key]) for key in self.priors}

317 318 319 320 321
        # Convert callable sampler_kwargs to strings to avoid pickling issues
        if dictionary.get('sampler_kwargs', None) is not None:
            for key in dictionary['sampler_kwargs']:
                if hasattr(dictionary['sampler_kwargs'][key], '__call__'):
                    dictionary['sampler_kwargs'][key] = str(dictionary['sampler_kwargs'])
322

Gregory Ashton's avatar
Fix #49  
Gregory Ashton committed
323
        try:
324
            deepdish.io.save(file_name, dictionary)
Gregory Ashton's avatar
Fix #49  
Gregory Ashton committed
325
        except Exception as e:
Gregory Ashton's avatar
Gregory Ashton committed
326
            logger.error("\n\n Saving the data has failed with the "
MoritzThomasHuebner's avatar
MoritzThomasHuebner committed
327
                         "following message:\n {} \n\n".format(e))
Gregory Ashton's avatar
Gregory Ashton committed
328

329
    def save_posterior_samples(self):
330
        """Saves posterior samples to a file"""
331
        filename = '{}/{}_posterior_samples.txt'.format(self.outdir, self.label)
332
        utils.check_directory_exists_and_if_not_mkdir(self.outdir)
333 334
        self.posterior.to_csv(filename, index=False, header=True)

Gregory Ashton's avatar
Gregory Ashton committed
335
    def get_latex_labels_from_parameter_keys(self, keys):
336 337 338 339 340 341 342 343 344 345 346 347 348
        """ Returns a list of latex_labels corresponding to the given keys

        Parameters
        ----------
        keys: list
            List of strings corresponding to the desired latex_labels

        Returns
        -------
        list: The desired latex_labels

        """
        latex_labels = []
Gregory Ashton's avatar
Gregory Ashton committed
349 350 351
        for k in keys:
            if k in self.search_parameter_keys:
                idx = self.search_parameter_keys.index(k)
352
                latex_labels.append(self.parameter_labels_with_unit[idx])
Gregory Ashton's avatar
Gregory Ashton committed
353
            elif k in self.parameter_labels:
354
                latex_labels.append(k)
Gregory Ashton's avatar
Gregory Ashton committed
355
            else:
Colm Talbot's avatar
Colm Talbot committed
356
                logger.debug(
357 358
                    'key {} not a parameter label or latex label'.format(k))
                latex_labels.append(' '.join(k.split('_')))
359
        return latex_labels
Gregory Ashton's avatar
Gregory Ashton committed
360

Gregory Ashton's avatar
Gregory Ashton committed
361 362 363 364 365 366 367 368 369
    @property
    def covariance_matrix(self):
        """ The covariance matrix of the samples the posterior """
        samples = self.posterior[self.search_parameter_keys].values
        return np.cov(samples.T)

    @property
    def posterior_volume(self):
        """ The posterior volume """
Gregory Ashton's avatar
Gregory Ashton committed
370 371 372
        if self.covariance_matrix.ndim == 0:
            return np.sqrt(self.covariance_matrix)
        else:
MoritzThomasHuebner's avatar
MoritzThomasHuebner committed
373 374
            return 1 / np.sqrt(np.abs(np.linalg.det(
                1 / self.covariance_matrix)))
Gregory Ashton's avatar
Gregory Ashton committed
375

376 377
    @staticmethod
    def prior_volume(priors):
Gregory Ashton's avatar
Gregory Ashton committed
378 379 380 381 382 383
        """ The prior volume, given a set of priors """
        return np.prod([priors[k].maximum - priors[k].minimum for k in priors])

    def occam_factor(self, priors):
        """ The Occam factor,

384 385 386
        See Chapter 28, `Mackay "Information Theory, Inference, and Learning
        Algorithms" <http://www.inference.org.uk/itprnn/book.html>`_ Cambridge
        University Press (2003).
Gregory Ashton's avatar
Gregory Ashton committed
387 388 389 390

        """
        return self.posterior_volume / self.prior_volume(priors)

391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406
    def get_one_dimensional_median_and_error_bar(self, key, fmt='.2f',
                                                 quantiles=[0.16, 0.84]):
        """ Calculate the median and error bar for a given key

        Parameters
        ----------
        key: str
            The parameter key for which to calculate the median and error bar
        fmt: str, ('.2f')
            A format string
        quantiles: list
            A length-2 list of the lower and upper-quantiles to calculate
            the errors bars for.

        Returns
        -------
407 408
        summary: namedtuple
            An object with attributes, median, lower, upper and string
409 410

        """
411 412
        summary = namedtuple('summary', ['median', 'lower', 'upper', 'string'])

413 414 415 416 417
        if len(quantiles) != 2:
            raise ValueError("quantiles must be of length 2")

        quants_to_compute = np.array([quantiles[0], 0.5, quantiles[1]])
        quants = np.percentile(self.posterior[key], quants_to_compute * 100)
418 419 420
        summary.median = quants[1]
        summary.plus = quants[2] - summary.median
        summary.minus = summary.median - quants[0]
421 422

        fmt = "{{0:{0}}}".format(fmt).format
423 424 425 426 427
        string_template = r"${{{0}}}_{{-{1}}}^{{+{2}}}$"
        summary.string = string_template.format(
            fmt(summary.median), fmt(summary.minus), fmt(summary.plus))
        return summary

Colm Talbot's avatar
Colm Talbot committed
428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474
    def plot_single_density(self, key, prior=None, cumulative=False,
                            title=None, truth=None, save=True,
                            file_base_name=None, bins=50, label_fontsize=16,
                            title_fontsize=16, quantiles=[0.16, 0.84], dpi=300):
        """ Plot a 1D marginal density, either probablility or cumulative.

        Parameters
        ----------
        key: str
            Name of the parameter to plot
        prior: {bool (True), bilby.core.prior.Prior}
            If true, add the stored prior probability density function to the
            one-dimensional marginal distributions. If instead a Prior
            is provided, this will be plotted.
        cumulative: bool
            If true plot the CDF
        title: bool
            If true, add 1D title of the median and (by default 1-sigma)
            error bars. To change the error bars, pass in the quantiles kwarg.
            See method `get_one_dimensional_median_and_error_bar` for further
            details). If `quantiles=None` is passed in, no title is added.
        truth: {bool, float}
            If true, plot self.injection_parameters[parameter].
            If float, plot this value.
        save: bool:
            If true, save plot to disk.
        file_base_name: str, optional
            If given, the base file name to use (by default `outdir/label_` is
            used)
        bins: int
            The number of histogram bins
        label_fontsize, title_fontsize: int
            The fontsizes for the labels and titles
        quantiles: list
            A length-2 list of the lower and upper-quantiles to calculate
            the errors bars for.
        dpi: int
            Dots per inch resolution of the plot

        Returns
        -------
        figure: matplotlib.pyplot.figure
            A matplotlib figure object
        """
        logger.info('Plotting {} marginal distribution'.format(key))
        label = self.get_latex_labels_from_parameter_keys([key])[0]
        fig, ax = plt.subplots()
475 476 477 478 479 480 481 482
        try:
            ax.hist(self.posterior[key].values, bins=bins, density=True,
                    histtype='step', cumulative=cumulative)
        except ValueError as e:
            logger.info(
                'Failed to generate 1d plot for {}, error message: {}'
                .format(key, e))
            return
Colm Talbot's avatar
Colm Talbot committed
483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504
        ax.set_xlabel(label, fontsize=label_fontsize)
        if truth is not None:
            ax.axvline(truth, ls='-', color='orange')

        summary = self.get_one_dimensional_median_and_error_bar(
            key, quantiles=quantiles)
        ax.axvline(summary.median - summary.minus, ls='--', color='C0')
        ax.axvline(summary.median + summary.plus, ls='--', color='C0')
        if title:
            ax.set_title(summary.string, fontsize=title_fontsize)

        if isinstance(prior, Prior):
            theta = np.linspace(ax.get_xlim()[0], ax.get_xlim()[1], 300)
            ax.plot(theta, Prior.prob(theta), color='C2')

        if save:
            fig.tight_layout()
            if cumulative:
                file_name = file_base_name + key + '_cdf'
            else:
                file_name = file_base_name + key + '_pdf'
            fig.savefig(file_name, dpi=dpi)
505 506 507
            plt.close(fig)
        else:
            return fig
Colm Talbot's avatar
Colm Talbot committed
508

509 510 511 512 513 514 515 516 517 518
    def plot_marginals(self, parameters=None, priors=None, titles=True,
                       file_base_name=None, bins=50, label_fontsize=16,
                       title_fontsize=16, quantiles=[0.16, 0.84], dpi=300):
        """ Plot 1D marginal distributions

        Parameters
        ----------
        parameters: (list, dict), optional
            If given, either a list of the parameter names to include, or a
            dictionary of parameter names and their "true" values to plot.
519
        priors: {bool (False), bilby.core.prior.PriorDict}
520
            If true, add the stored prior probability density functions to the
521
            one-dimensional marginal distributions. If instead a PriorDict
522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545
            is provided, this will be plotted.
        titles: bool
            If true, add 1D titles of the median and (by default 1-sigma)
            error bars. To change the error bars, pass in the quantiles kwarg.
            See method `get_one_dimensional_median_and_error_bar` for further
            details). If `quantiles=None` is passed in, no title is added.
        file_base_name: str, optional
            If given, the base file name to use (by default `outdir/label_` is
            used)
        bins: int
            The number of histogram bins
        label_fontsize, title_fontsize: int
            The fontsizes for the labels and titles
        quantiles: list
            A length-2 list of the lower and upper-quantiles to calculate
            the errors bars for.
        dpi: int
            Dots per inch resolution of the plot

        Returns
        -------
        """
        if isinstance(parameters, dict):
            plot_parameter_keys = list(parameters.keys())
Colm Talbot's avatar
Colm Talbot committed
546
            truths = parameters
547
        elif parameters is None:
Colm Talbot's avatar
Colm Talbot committed
548 549 550 551 552
            plot_parameter_keys = self.posterior.keys()
            if self.injection_parameters is None:
                truths = dict()
            else:
                truths = self.injection_parameters
553 554
        else:
            plot_parameter_keys = list(parameters)
Colm Talbot's avatar
Colm Talbot committed
555 556 557 558
            if self.injection_parameters is None:
                truths = dict()
            else:
                truths = self.injection_parameters
559 560

        if file_base_name is None:
Colm Talbot's avatar
Colm Talbot committed
561 562
            file_base_name = '{}/{}_1d/'.format(self.outdir, self.label)
            check_directory_exists_and_if_not_mkdir(file_base_name)
563 564

        if priors is True:
Colm Talbot's avatar
Colm Talbot committed
565 566
            priors = getattr(self, 'priors', dict())
        elif isinstance(priors, dict):
567
            pass
Colm Talbot's avatar
Colm Talbot committed
568 569
        elif priors in [False, None]:
            priors = dict()
570 571 572 573
        else:
            raise ValueError('Input priors={} not understood'.format(priors))

        for i, key in enumerate(plot_parameter_keys):
Colm Talbot's avatar
Colm Talbot committed
574 575
            if not isinstance(self.posterior[key].values[0], float):
                continue
Colm Talbot's avatar
Colm Talbot committed
576 577
            prior = priors.get(key, None)
            truth = truths.get(key, None)
Colm Talbot's avatar
Colm Talbot committed
578
            for cumulative in [False, True]:
579
                self.plot_single_density(
Colm Talbot's avatar
Colm Talbot committed
580 581
                    key, prior=prior, cumulative=cumulative, title=titles,
                    truth=truth, save=True, file_base_name=file_base_name,
Colm Talbot's avatar
Colm Talbot committed
582 583
                    bins=bins, label_fontsize=label_fontsize, dpi=dpi,
                    title_fontsize=title_fontsize, quantiles=quantiles)
584

585 586 587
    def plot_corner(self, parameters=None, priors=None, titles=True, save=True,
                    filename=None, dpi=300, **kwargs):
        """ Plot a corner-plot
Gregory Ashton's avatar
Gregory Ashton committed
588 589 590

        Parameters
        ----------
591 592 593
        parameters: (list, dict), optional
            If given, either a list of the parameter names to include, or a
            dictionary of parameter names and their "true" values to plot.
594
        priors: {bool (False), bilby.core.prior.PriorDict}
595
            If true, add the stored prior probability density functions to the
596
            one-dimensional marginal distributions. If instead a PriorDict
597
            is provided, this will be plotted.
598 599 600 601 602
        titles: bool
            If true, add 1D titles of the median and (by default 1-sigma)
            error bars. To change the error bars, pass in the quantiles kwarg.
            See method `get_one_dimensional_median_and_error_bar` for further
            details). If `quantiles=None` is passed in, no title is added.
603 604 605 606 607 608
        save: bool, optional
            If true, save the image using the given label and outdir
        filename: str, optional
            If given, overwrite the default filename
        dpi: int, optional
            Dots per inch resolution of the plot
609 610 611 612
        **kwargs:
            Other keyword arguments are passed to `corner.corner`. We set some
            defaults to improve the basic look and feel, but these can all be
            overridden.
Gregory Ashton's avatar
Gregory Ashton committed
613

614 615 616 617 618 619
        Notes
        -----
            The generation of the corner plot themselves is done by the corner
            python module, see https://corner.readthedocs.io for more
            information.

Gregory Ashton's avatar
Gregory Ashton committed
620 621 622 623
        Returns
        -------
        fig:
            A matplotlib figure instance
624

Gregory Ashton's avatar
Gregory Ashton committed
625
        """
626 627

        # If in testing mode, not corner plots are generated
628 629
        if utils.command_line_args.test:
            return
Gregory Ashton's avatar
Gregory Ashton committed
630

Colm Talbot's avatar
Colm Talbot committed
631
        # bilby default corner kwargs. Overwritten by anything passed to kwargs
632 633 634
        defaults_kwargs = dict(
            bins=50, smooth=0.9, label_kwargs=dict(fontsize=16),
            title_kwargs=dict(fontsize=16), color='#0072C1',
635
            truth_color='tab:orange', quantiles=[0.16, 0.84],
MoritzThomasHuebner's avatar
MoritzThomasHuebner committed
636
            levels=(1 - np.exp(-0.5), 1 - np.exp(-2), 1 - np.exp(-9 / 2.)),
637
            plot_density=False, plot_datapoints=True, fill_contours=True,
638 639 640 641 642 643
            max_n_ticks=3)

        if LooseVersion(matplotlib.__version__) < "2.1":
            defaults_kwargs['hist_kwargs'] = dict(normed=True)
        else:
            defaults_kwargs['hist_kwargs'] = dict(density=True)
644

645 646 647 648
        if 'lionize' in kwargs and kwargs['lionize'] is True:
            defaults_kwargs['truth_color'] = 'tab:blue'
            defaults_kwargs['color'] = '#FF8C00'

649 650 651
        defaults_kwargs.update(kwargs)
        kwargs = defaults_kwargs

652 653 654 655 656 657 658 659 660 661 662 663 664 665 666
        # Handle if truths was passed in
        if 'truth' in kwargs:
            kwargs['truths'] = kwargs.pop('truth')
        if kwargs.get('truths'):
            truths = kwargs.get('truths')
            if isinstance(parameters, list) and isinstance(truths, list):
                if len(parameters) != len(truths):
                    raise ValueError(
                        "Length of parameters and truths don't match")
            elif isinstance(truths, dict) and parameters is None:
                parameters = kwargs.pop('truths')
            else:
                raise ValueError(
                    "Combination of parameters and truths not understood")

Gregory Ashton's avatar
Gregory Ashton committed
667 668 669 670 671 672 673 674
        # If injection parameters where stored, use these as parameter values
        # but do not overwrite input parameters (or truths)
        cond1 = getattr(self, 'injection_parameters', None) is not None
        cond2 = parameters is None
        if cond1 and cond2:
            parameters = {key: self.injection_parameters[key] for key in
                          self.search_parameter_keys}

675 676 677 678 679 680 681 682 683
        # If parameters is a dictionary, use the keys to determine which
        # parameters to plot and the values as truths.
        if isinstance(parameters, dict):
            plot_parameter_keys = list(parameters.keys())
            kwargs['truths'] = list(parameters.values())
        elif parameters is None:
            plot_parameter_keys = self.search_parameter_keys
        else:
            plot_parameter_keys = list(parameters)
684

685
        # Get latex formatted strings for the plot labels
686 687
        kwargs['labels'] = kwargs.get(
            'labels', self.get_latex_labels_from_parameter_keys(
688
                plot_parameter_keys))
689

690 691 692 693
        # Unless already set, set the range to include all samples
        # This prevents ValueErrors being raised for parameters with no range
        kwargs['range'] = kwargs.get('range', [1] * len(plot_parameter_keys))

694 695
        # Create the data array to plot and pass everything to corner
        xs = self.posterior[plot_parameter_keys].values
696
        fig = corner.corner(xs, **kwargs)
697
        axes = fig.get_axes()
698 699 700

        #  Add the titles
        if titles and kwargs.get('quantiles', None) is not None:
701 702
            for i, par in enumerate(plot_parameter_keys):
                ax = axes[i + i * len(plot_parameter_keys)]
703 704
                if ax.title.get_text() == '':
                    ax.set_title(self.get_one_dimensional_median_and_error_bar(
705
                        par, quantiles=kwargs['quantiles']).string,
706 707 708
                        **kwargs['title_kwargs'])

        #  Add priors to the 1D plots
709 710 711
        if priors is True:
            priors = getattr(self, 'priors', False)
        if isinstance(priors, dict):
712 713
            for i, par in enumerate(plot_parameter_keys):
                ax = axes[i + i * len(plot_parameter_keys)]
714 715
                theta = np.linspace(ax.get_xlim()[0], ax.get_xlim()[1], 300)
                ax.plot(theta, priors[par].prob(theta), color='C2')
716 717 718 719
        elif priors in [False, None]:
            pass
        else:
            raise ValueError('Input priors={} not understood'.format(priors))
720

721
        if save:
722 723 724
            if filename is None:
                utils.check_directory_exists_and_if_not_mkdir(self.outdir)
                filename = '{}/{}_corner.png'.format(self.outdir, self.label)
Gregory Ashton's avatar
Gregory Ashton committed
725
            logger.debug('Saving corner plot to {}'.format(filename))
726
            fig.savefig(filename, dpi=dpi)
727
            plt.close(fig)
728

729
        return fig
730

Gregory Ashton's avatar
Gregory Ashton committed
731
    def plot_walkers(self, **kwargs):
MoritzThomasHuebner's avatar
A typo  
MoritzThomasHuebner committed
732
        """ Method to plot the trace of the walkers in an ensemble MCMC plot """
733
        if hasattr(self, 'walkers') is False:
Gregory Ashton's avatar
Gregory Ashton committed
734
            logger.warning("Cannot plot_walkers as no walkers are saved")
735
            return
736 737 738

        if utils.command_line_args.test:
            return
739 740 741

        nwalkers, nsteps, ndim = self.walkers.shape
        idxs = np.arange(nsteps)
MoritzThomasHuebner's avatar
MoritzThomasHuebner committed
742
        fig, axes = plt.subplots(nrows=ndim, figsize=(6, 3 * ndim))
743 744
        walkers = self.walkers[:, :, :]
        for i, ax in enumerate(axes):
MoritzThomasHuebner's avatar
MoritzThomasHuebner committed
745
            ax.plot(idxs[:self.nburn + 1], walkers[:, :self.nburn + 1, i].T,
746 747 748 749 750 751 752 753 754 755
                    lw=0.1, color='r')
            ax.set_ylabel(self.parameter_labels[i])

        for i, ax in enumerate(axes):
            ax.plot(idxs[self.nburn:], walkers[:, self.nburn:, i].T, lw=0.1,
                    color='k')
            ax.set_ylabel(self.parameter_labels[i])

        fig.tight_layout()
        filename = '{}/{}_walkers.png'.format(self.outdir, self.label)
Gregory Ashton's avatar
Gregory Ashton committed
756
        logger.debug('Saving walkers plot to {}'.format('filename'))
757
        utils.check_directory_exists_and_if_not_mkdir(self.outdir)
758
        fig.savefig(filename)
759
        plt.close(fig)
760

761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791
    def plot_with_data(self, model, x, y, ndraws=1000, npoints=1000,
                       xlabel=None, ylabel=None, data_label='data',
                       data_fmt='o', draws_label=None, filename=None,
                       maxl_label='max likelihood', dpi=300):
        """ Generate a figure showing the data and fits to the data

        Parameters
        ----------
        model: function
            A python function which when called as `model(x, **kwargs)` returns
            the model prediction (here `kwargs` is a dictionary of key-value
            pairs of the model parameters.
        x, y: np.ndarray
            The independent and dependent data to plot
        ndraws: int
            Number of draws from the posterior to plot
        npoints: int
            Number of points used to plot the smoothed fit to the data
        xlabel, ylabel: str
            Labels for the axes
        data_label, draws_label, maxl_label: str
            Label for the data, draws, and max likelihood legend
        data_fmt: str
            Matpltolib fmt code, defaults to `'-o'`
        dpi: int
            Passed to `plt.savefig`
        filename: str
            If given, the filename to use. Otherwise, the filename is generated
            from the outdir and label attributes.

        """
792 793 794 795 796 797

        # Determine model_posterior, the subset of the full posterior which
        # should be passed into the model
        model_keys = infer_parameters_from_function(model)
        model_posterior = self.posterior[model_keys]

798 799 800 801
        xsmooth = np.linspace(np.min(x), np.max(x), npoints)
        fig, ax = plt.subplots()
        logger.info('Plotting {} draws'.format(ndraws))
        for _ in range(ndraws):
802
            s = model_posterior.sample().to_dict('records')[0]
803 804
            ax.plot(xsmooth, model(xsmooth, **s), alpha=0.25, lw=0.1, color='r',
                    label=draws_label)
805 806 807
        try:
            if all(~np.isnan(self.posterior.log_likelihood)):
                logger.info('Plotting maximum likelihood')
Colm Talbot's avatar
Colm Talbot committed
808
                s = model_posterior.iloc[self.posterior.log_likelihood.idxmax()]
809 810
                ax.plot(xsmooth, model(xsmooth, **s), lw=1, color='k',
                        label=maxl_label)
811
        except (AttributeError, TypeError):
812 813
            logger.debug(
                "No log likelihood values stored, unable to plot max")
814 815 816 817 818 819 820 821 822 823 824 825 826 827

        ax.plot(x, y, data_fmt, markersize=2, label=data_label)

        if xlabel is not None:
            ax.set_xlabel(xlabel)
        if ylabel is not None:
            ax.set_ylabel(ylabel)

        handles, labels = plt.gca().get_legend_handles_labels()
        by_label = OrderedDict(zip(labels, handles))
        plt.legend(by_label.values(), by_label.keys())
        ax.legend(numpoints=3)
        fig.tight_layout()
        if filename is None:
828
            utils.check_directory_exists_and_if_not_mkdir(self.outdir)
829 830
            filename = '{}/{}_plot_with_data'.format(self.outdir, self.label)
        fig.savefig(filename, dpi=dpi)
831
        plt.close(fig)
832

833 834
    def samples_to_posterior(self, likelihood=None, priors=None,
                             conversion_function=None):
835
        """
836 837 838
        Convert array of samples to posterior (a Pandas data frame)

        Also applies the conversion function to any stored posterior
839

840 841
        Parameters
        ----------
Colm Talbot's avatar
Colm Talbot committed
842
        likelihood: bilby.likelihood.GravitationalWaveTransient, optional
843 844
            GravitationalWaveTransient likelihood used for sampling.
        priors: dict, optional
845
            Dictionary of prior object, used to fill in delta function priors.
846
        conversion_function: function, optional
847 848
            Function which adds in extra parameters to the data frame,
            should take the data_frame, likelihood and prior as arguments.
849
        """
850 851 852
        try:
            data_frame = self.posterior
        except ValueError:
853 854
            data_frame = pd.DataFrame(
                self.samples, columns=self.search_parameter_keys)
855
            for key in priors:
856
                if isinstance(priors[key], DeltaFunction):
857
                    data_frame[key] = priors[key].peak
858 859 860 861
                elif isinstance(priors[key], float):
                    data_frame[key] = priors[key]
            data_frame['log_likelihood'] = getattr(
                self, 'log_likelihood_evaluations', np.nan)
862 863 864 865 866
            if self.log_prior_evaluations is None:
                data_frame['log_prior'] = self.priors.ln_prob(
                    data_frame[self.search_parameter_keys], axis=0)
            else:
                data_frame['log_prior'] = self.log_prior_evaluations
867
        if conversion_function is not None:
868
            data_frame = conversion_function(data_frame, likelihood, priors)
869
        self.posterior = data_frame
Gregory Ashton's avatar
Gregory Ashton committed
870

Colm Talbot's avatar
Colm Talbot committed
871
    def calculate_prior_values(self, priors):
872 873 874 875 876
        """
        Evaluate prior probability for each parameter for each sample.

        Parameters
        ----------
877
        priors: dict, PriorDict
878 879 880 881 882 883 884 885 886 887 888
            Prior distributions
        """
        self.prior_values = pd.DataFrame()
        for key in priors:
            if key in self.posterior.keys():
                if isinstance(priors[key], DeltaFunction):
                    continue
                else:
                    self.prior_values[key]\
                        = priors[key].prob(self.posterior[key].values)

Colm Talbot's avatar
Colm Talbot committed
889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931
    def get_all_injection_credible_levels(self):
        """
        Get credible levels for all parameters in self.injection_parameters

        Returns
        -------
        credible_levels: dict
            The credible levels at which the injected parameters are found.
        """
        if self.injection_parameters is None:
            raise(TypeError, "Result object has no 'injection_parameters'. "
                             "Cannot copmute credible levels.")
        credible_levels = {key: self.get_injection_credible_level(key)
                           for key in self.search_parameter_keys
                           if isinstance(self.injection_parameters[key], float)}
        return credible_levels

    def get_injection_credible_level(self, parameter):
        """
        Get the credible level of the injected parameter

        Calculated as CDF(injection value)

        Parameters
        ----------
        parameter: str
            Parameter to get credible level for
        Returns
        -------
        float: credible level
        """
        if self.injection_parameters is None:
            raise(TypeError, "Result object has no 'injection_parameters'. "
                             "Cannot copmute credible levels.")
        if parameter in self.posterior and\
                parameter in self.injection_parameters:
            credible_level =\
                sum(self.posterior[parameter].values <
                    self.injection_parameters[parameter]) / len(self.posterior)
            return credible_level
        else:
            return np.nan

932
    def _check_attribute_match_to_other_object(self, name, other_object):
933 934 935 936 937 938 939 940 941 942 943 944 945 946
        """ Check attribute name exists in other_object and is the same

        Parameters
        ----------
        name: str
            Name of the attribute in this instance
        other_object: object
            Other object with attributes to compare with

        Returns
        -------
        bool: True if attribute name matches with an attribute of other_object, False otherwise

        """
947
        A = getattr(self, name, False)
948
        B = getattr(other_object, name, False)
Gregory Ashton's avatar
Gregory Ashton committed
949
        logger.debug('Checking {} value: {}=={}'.format(name, A, B))
950 951 952 953 954
        if (A is not False) and (B is not False):
            typeA = type(A)
            typeB = type(B)
            if typeA == typeB:
                if typeA in [str, float, int, dict, list]:
Gregory Ashton's avatar
Gregory Ashton committed
955 956 957 958
                    try:
                        return A == B
                    except ValueError:
                        return False
959 960 961
                elif typeA in [np.ndarray]:
                    return np.all(A == B)
        return False
962

963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000
    @property
    def kde(self):
        """ Kernel density estimate built from the stored posterior

        Uses `scipy.stats.gaussian_kde` to generate the kernel density
        """
        try:
            return self._kde
        except AttributeError:
            self._kde = scipy.stats.gaussian_kde(
                self.posterior[self.search_parameter_keys].values.T)
            return self._kde

    def posterior_probability(self, sample):
        """ Calculate the posterior probabily for a new sample

        This queries a Kernel Density Estimate of the posterior to calculate
        the posterior probability density for the new sample.

        Parameters
        ----------
        sample: dict, or list of dictionaries
            A dictionary containing all the keys from
            self.search_parameter_keys and corresponding values at which to
            calculate the posterior probability

        Returns
        -------
        p: array-like,
            The posterior probability of the sample

        """
        if isinstance(sample, dict):
            sample = [sample]
        ordered_sample = [[s[key] for key in self.search_parameter_keys]
                          for s in sample]
        return self.kde(ordered_sample)

1001 1002

def plot_multiple(results, filename=None, labels=None, colours=None,
1003
                  save=True, evidences=False, **kwargs):
1004 1005 1006 1007 1008
    """ Generate a corner plot overlaying two sets of results

    Parameters
    ----------
    results: list
Colm Talbot's avatar
Colm Talbot committed
1009
        A list of `bilby.core.result.Result` objects containing the samples to
1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025
        plot.
    filename: str
        File name to save the figure to. If None (default), a filename is
        constructed from the outdir of the first element of results and then
        the labels for all the result files.
    labels: list
        List of strings to use when generating a legend. If None (default), the
        `label` attribute of each result in `results` is used.
    colours: list
        The colours for each result. If None, default styles are applied.
    save: bool
        If true, save the figure
    kwargs: dict
        All other keyword arguments are passed to `result.plot_corner`.
        However, `show_titles` and `truths` are ignored since they would be
        ambiguous on such a plot.
1026 1027 1028
    evidences: bool, optional
        Add the log-evidence calculations to the legend. If available, the
        Bayes factor will be used instead.
1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040

    Returns
    -------
    fig:
        A matplotlib figure instance

    """

    kwargs['show_titles'] = False
    kwargs['truths'] = None

    fig = results[0].plot_corner(save=False, **kwargs)
Gregory Ashton's avatar
Gregory Ashton committed
1041
    default_filename = '{}/{}'.format(results[0].outdir, 'combined')
1042 1043 1044 1045 1046 1047 1048
    lines = []
    default_labels = []
    for i, result in enumerate(results):
        if colours:
            c = colours[i]
        else:
            c = 'C{}'.format(i)
Gregory Ashton's avatar
Gregory Ashton committed
1049 1050
        hist_kwargs = kwargs.get('hist_kwargs', dict())
        hist_kwargs['color'] = c
1051 1052 1053 1054 1055
        fig = result.plot_corner(fig=fig, save=False, color=c, **kwargs)
        default_filename += '_{}'.format(result.label)
        lines.append(matplotlib.lines.Line2D([0], [0], color=c))
        default_labels.append(result.label)

Gregory Ashton's avatar
Gregory Ashton committed
1056 1057 1058 1059 1060
    # Rescale the axes
    for i, ax in enumerate(fig.axes):
        ax.autoscale()
    plt.draw()

1061 1062 1063
    if labels is None:
        labels = default_labels

1064 1065
    if evidences:
        if np.isnan(results[0].log_bayes_factor):
1066
            template = ' $\mathrm{{ln}}(Z)={lnz:1.3g}$'
1067
        else:
1068 1069 1070 1071
            template = ' $\mathrm{{ln}}(B)={lnbf:1.3g}$'
        labels = [template.format(lnz=result.log_evidence,
                                  lnbf=result.log_bayes_factor)
                  for ii, result in enumerate(results)]
1072

1073 1074
    axes = fig.get_axes()
    ndim = int(np.sqrt(len(axes)))
MoritzThomasHuebner's avatar
MoritzThomasHuebner committed
1075
    axes[ndim - 1].legend(lines, labels)
1076 1077 1078 1079 1080 1081 1082

    if filename is None:
        filename = default_filename

    if save:
        fig.savefig(filename)
    return fig
Colm Talbot's avatar
Colm Talbot committed
1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123