mcmc_based_searches.py 87.9 KB
Newer Older
Gregory Ashton's avatar
Gregory Ashton committed
1
2
""" Searches using MCMC-based methods """

3
import sys
Gregory Ashton's avatar
Gregory Ashton committed
4
import os
5
import copy
Gregory Ashton's avatar
Gregory Ashton committed
6
import logging
7
from collections import OrderedDict
8
9
10
11
12
13
14
15

import numpy as np
import matplotlib
import matplotlib.pyplot as plt
import emcee
import corner
import dill as pickle

16
import core
Gregory Ashton's avatar
Gregory Ashton committed
17
from core import tqdm, args, earth_ephem, sun_ephem
18
from optimal_setup_functions import get_V_estimate
Gregory Ashton's avatar
Gregory Ashton committed
19
20
from optimal_setup_functions import get_optimal_setup
import helper_functions
21
22


23
class MCMCSearch(core.BaseSearchClass):
Gregory Ashton's avatar
Gregory Ashton committed
24
    """ MCMC search using ComputeFstat"""
25
26

    symbol_dictionary = dict(
27
        F0='$f$', F1='$\dot{f}$', F2='$\ddot{f}$', Alpha=r'$\alpha$',
28
29
        Delta='$\delta$', asini='asini', period='P', ecc='ecc', tp='tp',
        argp='argp')
30
    unit_dictionary = dict(
31
32
        F0='Hz', F1='Hz/s', F2='Hz/s$^2$', Alpha=r'rad', Delta='rad',
        asini='', period='s', ecc='', tp='', argp='')
33
34
35
    rescale_dictionary = {}


Gregory Ashton's avatar
Gregory Ashton committed
36
    @helper_functions.initializer
Gregory Ashton's avatar
Gregory Ashton committed
37
38
    def __init__(self, label, outdir, theta_prior, tref, minStartTime,
                 maxStartTime, sftfilepath=None, nsteps=[100, 100],
39
40
                 nwalkers=100, ntemps=1, log10temperature_min=-5,
                 theta_initial=None, scatter_val=1e-10,
41
                 binary=False, BSGL=False, minCoverFreq=None,
42
                 maxCoverFreq=None, detectors=None, earth_ephem=None,
43
                 sun_ephem=None, injectSources=None, assumeSqrtSX=None):
44
45
46
47
        """
        Parameters
        label, outdir: str
            A label and directory to read/write data from/to
48
        sftfilepath: str
49
50
            Pattern to match SFTs using wildcards (*?) and ranges [0-9];
            mutiple patterns can be given separated by colons.
51
        theta_prior: dict
52
53
54
55
            Dictionary of priors and fixed values for the search parameters.
            For each parameters (key of the dict), if it is to be held fixed
            the value should be the constant float, if it is be searched, the
            value should be a dictionary of the prior.
56
57
58
59
        theta_initial: dict, array, (None)
            Either a dictionary of distribution about which to distribute the
            initial walkers about, an array (from which the walkers will be
            scattered by scatter_val, or  None in which case the prior is used.
60
        tref, minStartTime, maxStartTime: int
61
62
63
64
65
66
            GPS seconds of the reference time, start time and end time
        nsteps: list (m,)
            List specifying the number of steps to take, the last two entries
            give the nburn and nprod of the 'production' run, all entries
            before are for iterative initialisation steps (usually just one)
            e.g. [1000, 1000, 500].
67
68
69
70
71
72
73
74
        nwalkers, ntemps: int,
            The number of walkers and temperates to use in the parallel
            tempered PTSampler.
        log10temperature_min float < 0
            The  log_10(tmin) value, the set of betas passed to PTSampler are
            generated from np.logspace(0, log10temperature_min, ntemps).
        binary: Bool
            If true, search over binary parameters
75
        detectors: str
76
77
            Two character reference to the data to use, specify None for no
            contraint.
78
79
80
81
82
83
84
85
86
87
        minCoverFreq, maxCoverFreq: float
            Minimum and maximum instantaneous frequency which will be covered
            over the SFT time span as passed to CreateFstatInput
        earth_ephem, sun_ephem: str
            Paths of the two files containing positions of Earth and Sun,
            respectively at evenly spaced times, as passed to CreateFstatInput
            If None defaults defined in BaseSearchClass will be used

        """

Gregory Ashton's avatar
Gregory Ashton committed
88
89
        if os.path.isdir(outdir) is False:
            os.mkdir(outdir)
90
        self._add_log_file()
Gregory Ashton's avatar
Gregory Ashton committed
91
92
        logging.info(
            'Set-up MCMC search for model {} on data {}'.format(
93
                self.label, self.sftfilepath))
94
        self.pickle_path = '{}/{}_saved_data.p'.format(self.outdir, self.label)
95
        self._unpack_input_theta()
96
        self.ndim = len(self.theta_keys)
97
98
99
100
        if self.log10temperature_min:
            self.betas = np.logspace(0, self.log10temperature_min, self.ntemps)
        else:
            self.betas = None
101

102
103
104
105
106
107
108
109
        if earth_ephem is None:
            self.earth_ephem = self.earth_ephem_default
        if sun_ephem is None:
            self.sun_ephem = self.sun_ephem_default

        if args.clean and os.path.isfile(self.pickle_path):
            os.rename(self.pickle_path, self.pickle_path+".old")

110
        self._log_input()
111

112
    def _log_input(self):
113
        logging.info('theta_prior = {}'.format(self.theta_prior))
114
        logging.info('nwalkers={}'.format(self.nwalkers))
115
116
117
118
        logging.info('scatter_val = {}'.format(self.scatter_val))
        logging.info('nsteps = {}'.format(self.nsteps))
        logging.info('ntemps = {}'.format(self.ntemps))
        logging.info('log10temperature_min = {}'.format(
119
            self.log10temperature_min))
120

121
    def _initiate_search_object(self):
122
        logging.info('Setting up search object')
123
        self.search = core.ComputeFstat(
124
125
126
            tref=self.tref, sftfilepath=self.sftfilepath,
            minCoverFreq=self.minCoverFreq, maxCoverFreq=self.maxCoverFreq,
            earth_ephem=self.earth_ephem, sun_ephem=self.sun_ephem,
127
            detectors=self.detectors, BSGL=self.BSGL, transient=False,
128
            minStartTime=self.minStartTime, maxStartTime=self.maxStartTime,
129
130
            binary=self.binary, injectSources=self.injectSources,
            assumeSqrtSX=self.assumeSqrtSX)
131
132

    def logp(self, theta_vals, theta_prior, theta_keys, search):
133
        H = [self._generic_lnprior(**theta_prior[key])(p) for p, key in
134
135
136
137
138
139
             zip(theta_vals, theta_keys)]
        return np.sum(H)

    def logl(self, theta, search):
        for j, theta_i in enumerate(self.theta_idxs):
            self.fixed_theta[theta_i] = theta[j]
140
141
        FS = search.compute_fullycoherent_det_stat_single_point(
            *self.fixed_theta)
142
143
        return FS

144
    def _unpack_input_theta(self):
145
        full_theta_keys = ['F0', 'F1', 'F2', 'Alpha', 'Delta']
146
147
148
        if self.binary:
            full_theta_keys += [
                'asini', 'period', 'ecc', 'tp', 'argp']
149
150
        full_theta_keys_copy = copy.copy(full_theta_keys)

151
152
        full_theta_symbols = ['$f$', '$\dot{f}$', '$\ddot{f}$', r'$\alpha$',
                              r'$\delta$']
153
154
        if self.binary:
            full_theta_symbols += [
155
                'asini', 'period', 'ecc', 'tp', 'argp']
156

157
158
        self.theta_keys = []
        fixed_theta_dict = {}
159
        for key, val in self.theta_prior.iteritems():
160
161
            if type(val) is dict:
                fixed_theta_dict[key] = 0
Gregory Ashton's avatar
Gregory Ashton committed
162
                self.theta_keys.append(key)
163
164
165
166
167
168
            elif type(val) in [float, int, np.float64]:
                fixed_theta_dict[key] = val
            else:
                raise ValueError(
                    'Type {} of {} in theta not recognised'.format(
                        type(val), key))
Gregory Ashton's avatar
Gregory Ashton committed
169
            full_theta_keys_copy.pop(full_theta_keys_copy.index(key))
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184

        if len(full_theta_keys_copy) > 0:
            raise ValueError(('Input dictionary `theta` is missing the'
                              'following keys: {}').format(
                                  full_theta_keys_copy))

        self.fixed_theta = [fixed_theta_dict[key] for key in full_theta_keys]
        self.theta_idxs = [full_theta_keys.index(k) for k in self.theta_keys]
        self.theta_symbols = [full_theta_symbols[i] for i in self.theta_idxs]

        idxs = np.argsort(self.theta_idxs)
        self.theta_idxs = [self.theta_idxs[i] for i in idxs]
        self.theta_symbols = [self.theta_symbols[i] for i in idxs]
        self.theta_keys = [self.theta_keys[i] for i in idxs]

185
    def _check_initial_points(self, p0):
186
187
188
189
190
191
192
193
194
195
196
197
198
        for nt in range(self.ntemps):
            logging.info('Checking temperature {} chains'.format(nt))
            initial_priors = np.array([
                self.logp(p, self.theta_prior, self.theta_keys, self.search)
                for p in p0[nt]])
            number_of_initial_out_of_bounds = sum(initial_priors == -np.inf)

            if number_of_initial_out_of_bounds > 0:
                logging.warning(
                    'Of {} initial values, {} are -np.inf due to the prior'
                    .format(len(initial_priors),
                            number_of_initial_out_of_bounds))

199
                p0 = self._generate_new_p0_to_fix_initial_points(
200
201
                    p0, nt, initial_priors)

202
    def _generate_new_p0_to_fix_initial_points(self, p0, nt, initial_priors):
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
        logging.info('Attempting to correct intial values')
        idxs = np.arange(self.nwalkers)[initial_priors == -np.inf]
        count = 0
        while sum(initial_priors == -np.inf) > 0 and count < 100:
            for j in idxs:
                p0[nt][j] = (p0[nt][np.random.randint(0, self.nwalkers)]*(
                             1+np.random.normal(0, 1e-10, self.ndim)))
            initial_priors = np.array([
                self.logp(p, self.theta_prior, self.theta_keys,
                          self.search)
                for p in p0[nt]])
            count += 1

        if sum(initial_priors == -np.inf) > 0:
            logging.info('Failed to fix initial priors')
        else:
            logging.info('Suceeded to fix initial priors')

        return p0
222

223
    def _OLD_run_sampler_with_progress_bar(self, sampler, ns, p0):
224
225
        for result in tqdm(sampler.sample(p0, iterations=ns), total=ns):
            pass
Gregory Ashton's avatar
Gregory Ashton committed
226
227
        return sampler

228
229
    def setup_convergence_testing(
            self, convergence_period=10, convergence_length=10,
230
            convergence_burnin_fraction=0.25, convergence_threshold_number=10,
231
232
            convergence_threshold=1.2, convergence_prod_threshold=2,
            convergence_plot_upper_lim=2):
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
258
        """
        If called, convergence testing is used during the MCMC simulation

        This uses the Gelmanr-Rubin statistic based on the ratio of between and
        within walkers variance. The original statistic was developed for
        multiple (independent) MCMC simulations, in this context we simply use
        the walkers

        Parameters
        ----------
        convergence_period: int
            period (in number of steps) at which to test convergence
        convergence_length: int
            number of steps to use in testing convergence - this should be
            large enough to measure the variance, but if it is too long
            this will result in incorect early convergence tests
        convergence_burnin_fraction: float [0, 1]
            the fraction of the burn-in period after which to start testing
        convergence_threshold_number: int
            the number of consecutive times where the test passes after which
            to break the burn-in and go to production
        convergence_threshold: float
            the threshold to use in diagnosing convergence. Gelman & Rubin
            recomend a value of 1.2, 1.1 for strict convergence
        convergence_prod_threshold: float
            the threshold to test the production values with
259
260
        convergence_plot_upper_lim: float
            the upper limit to use in the diagnostic plot
261
        """
262
263
264
265
266
267
268

        if convergence_length > convergence_period:
            raise ValueError('convergence_length must be < convergence_period')
        logging.info('Setting up convergence testing')
        self.convergence_length = convergence_length
        self.convergence_period = convergence_period
        self.convergence_burnin_fraction = convergence_burnin_fraction
269
        self.convergence_prod_threshold = convergence_prod_threshold
270
271
272
273
274
        self.convergence_diagnostic = []
        self.convergence_diagnosticx = []
        self.convergence_threshold_number = convergence_threshold_number
        self.convergence_threshold = convergence_threshold
        self.convergence_number = 0
275
        self.convergence_plot_upper_lim = convergence_plot_upper_lim
276

277
    def _get_convergence_statistic(self, i, sampler):
278
279
280
281
282
283
284
285
286
        s = sampler.chain[0, :, i-self.convergence_length+1:i+1, :]
        within_std = np.mean(np.var(s, axis=1), axis=0)
        per_walker_mean = np.mean(s, axis=1)
        mean = np.mean(per_walker_mean, axis=0)
        between_std = np.sqrt(np.mean((per_walker_mean-mean)**2, axis=0))
        W = within_std
        B_over_n = between_std**2 / self.convergence_period
        Vhat = ((self.convergence_period-1.)/self.convergence_period * W
                + B_over_n + B_over_n / float(self.nwalkers))
287
        c = np.sqrt(Vhat/W)
288
        self.convergence_diagnostic.append(c)
289
        self.convergence_diagnosticx.append(i - self.convergence_length/2)
290
291
        return c

292
    def _burnin_convergence_test(self, i, sampler, nburn):
293
294
        if i < self.convergence_burnin_fraction*nburn:
            return False
295
        if np.mod(i+1, self.convergence_period) != 0:
296
            return False
297
        c = self._get_convergence_statistic(i, sampler)
298
299
        if np.all(c < self.convergence_threshold):
            self.convergence_number += 1
300
301
        else:
            self.convergence_number = 0
302
303
        return self.convergence_number > self.convergence_threshold_number

304
    def _prod_convergence_test(self, i, sampler, nburn):
305
306
307
        testA = i > nburn + self.convergence_length
        testB = np.mod(i+1, self.convergence_period) == 0
        if testA and testB:
308
            self._get_convergence_statistic(i, sampler)
309

310
    def _check_production_convergence(self, k):
311
312
313
314
315
316
317
318
        bools = np.any(
            np.array(self.convergence_diagnostic)[k:, :]
            > self.convergence_prod_threshold, axis=1)
        if np.any(bools):
            logging.warning(
                '{} convergence tests in the production run of {} failed'
                .format(np.sum(bools), len(bools)))

319
    def _run_sampler(self, sampler, p0, nprod=0, nburn=0):
320
        if hasattr(self, 'convergence_period'):
321
322
323
324
            logging.info('Running {} burn-in steps with convergence testing'
                         .format(nburn))
            iterator = tqdm(sampler.sample(p0, iterations=nburn), total=nburn)
            for i, output in enumerate(iterator):
325
                if self._burnin_convergence_test(i, sampler, nburn):
326
327
328
329
330
331
332
333
334
335
336
                    logging.info(
                        'Converged at {} before max number {} of steps reached'
                        .format(i, nburn))
                    self.convergence_idx = i
                    break
            iterator.close()
            logging.info('Running {} production steps'.format(nprod))
            j = nburn
            k = len(self.convergence_diagnostic)
            for result in tqdm(sampler.sample(output[0], iterations=nprod),
                               total=nprod):
337
                self._prod_convergence_test(j, sampler, nburn)
338
                j += 1
339
            self._check_production_convergence(k)
340
341
342
343
344
345
            return sampler
        else:
            for result in tqdm(sampler.sample(p0, iterations=nburn+nprod),
                               total=nburn+nprod):
                pass
            return sampler
346

347
    def run(self, proposal_scale_factor=2, create_plots=True, **kwargs):
348
        """ Run the MCMC simulatation """
349

350
        self.old_data_is_okay_to_use = self._check_old_data_is_okay_to_use()
351
352
353
        if self.old_data_is_okay_to_use is True:
            logging.warning('Using saved data from {}'.format(
                self.pickle_path))
354
            d = self.get_saved_data_dictionary()
355
356
357
358
359
            self.samples = d['samples']
            self.lnprobs = d['lnprobs']
            self.lnlikes = d['lnlikes']
            return

360
        self._initiate_search_object()
361
362
363
364

        sampler = emcee.PTSampler(
            self.ntemps, self.nwalkers, self.ndim, self.logl, self.logp,
            logpargs=(self.theta_prior, self.theta_keys, self.search),
365
            loglargs=(self.search,), betas=self.betas, a=proposal_scale_factor)
366

367
368
369
        p0 = self._generate_initial_p0()
        p0 = self._apply_corrections_to_p0(p0)
        self._check_initial_points(p0)
370
371
372
373

        ninit_steps = len(self.nsteps) - 2
        for j, n in enumerate(self.nsteps[:-2]):
            logging.info('Running {}/{} initialisation with {} steps'.format(
Gregory Ashton's avatar
Gregory Ashton committed
374
                j, ninit_steps, n))
375
            sampler = self._run_sampler(sampler, p0, nburn=n)
376
377
            logging.info("Mean acceptance fraction: {}"
                         .format(np.mean(sampler.acceptance_fraction, axis=1)))
378
379
380
            if self.ntemps > 1:
                logging.info("Tswap acceptance fraction: {}"
                             .format(sampler.tswap_acceptance_fraction))
381
            if create_plots:
382
                fig, axes = self._plot_walkers(sampler,
383
384
385
386
                                              symbols=self.theta_symbols,
                                              **kwargs)
                fig.tight_layout()
                fig.savefig('{}/{}_init_{}_walkers.png'.format(
387
                    self.outdir, self.label, j), dpi=400)
388

389
390
391
            p0 = self._get_new_p0(sampler)
            p0 = self._apply_corrections_to_p0(p0)
            self._check_initial_points(p0)
392
393
            sampler.reset()

Gregory Ashton's avatar
Gregory Ashton committed
394
395
396
397
        if len(self.nsteps) > 1:
            nburn = self.nsteps[-2]
        else:
            nburn = 0
398
399
400
        nprod = self.nsteps[-1]
        logging.info('Running final burn and prod with {} steps'.format(
            nburn+nprod))
401
        sampler = self._run_sampler(sampler, p0, nburn=nburn, nprod=nprod)
402
403
        logging.info("Mean acceptance fraction: {}"
                     .format(np.mean(sampler.acceptance_fraction, axis=1)))
404
405
406
        if self.ntemps > 1:
            logging.info("Tswap acceptance fraction: {}"
                         .format(sampler.tswap_acceptance_fraction))
407

408
        if create_plots:
409
            fig, axes = self._plot_walkers(sampler, symbols=self.theta_symbols,
410
                                          nprod=nprod, **kwargs)
411
412
413
            fig.tight_layout()
            fig.savefig('{}/{}_walkers.png'.format(self.outdir, self.label),
                        dpi=200)
414
415
416
417

        samples = sampler.chain[0, :, nburn:, :].reshape((-1, self.ndim))
        lnprobs = sampler.lnprobability[0, :, nburn:].reshape((-1))
        lnlikes = sampler.lnlikelihood[0, :, nburn:].reshape((-1))
418
        all_lnlikelihood = sampler.lnlikelihood
419
420
421
        self.samples = samples
        self.lnprobs = lnprobs
        self.lnlikes = lnlikes
422
423
        self.all_lnlikelihood = all_lnlikelihood
        self._save_data(sampler, samples, lnprobs, lnlikes, all_lnlikelihood)
424

425
    def _get_rescale_multiplier_for_key(self, key):
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
        """ Get the rescale multiplier from the rescale_dictionary

        Can either be a float, a string (in which case it is interpretted as
        a attribute of the MCMCSearch class, e.g. minStartTime, or non-existent
        in which case 0 is returned
        """
        if key not in self.rescale_dictionary:
            return 1

        if 'multiplier' in self.rescale_dictionary[key]:
            val = self.rescale_dictionary[key]['multiplier']
            if type(val) == str:
                if hasattr(self, val):
                    multiplier = getattr(
                        self, self.rescale_dictionary[key]['multiplier'])
                else:
                    raise ValueError(
                        "multiplier {} not a class attribute".format(val))
            else:
                multiplier = val
        else:
            multiplier = 1
        return multiplier

450
    def _get_rescale_subtractor_for_key(self, key):
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
        """ Get the rescale subtractor from the rescale_dictionary

        Can either be a float, a string (in which case it is interpretted as
        a attribute of the MCMCSearch class, e.g. minStartTime, or non-existent
        in which case 0 is returned
        """
        if key not in self.rescale_dictionary:
            return 0

        if 'subtractor' in self.rescale_dictionary[key]:
            val = self.rescale_dictionary[key]['subtractor']
            if type(val) == str:
                if hasattr(self, val):
                    subtractor = getattr(
                        self, self.rescale_dictionary[key]['subtractor'])
                else:
                    raise ValueError(
                        "subtractor {} not a class attribute".format(val))
            else:
                subtractor = val
        else:
            subtractor = 0
        return subtractor

475
    def _scale_samples(self, samples, theta_keys):
476
        """ Scale the samples using the rescale_dictionary """
477
478
479
480
        for key in theta_keys:
            if key in self.rescale_dictionary:
                idx = theta_keys.index(key)
                s = samples[:, idx]
481
                subtractor = self._get_rescale_subtractor_for_key(key)
482
                s = s - subtractor
483
                multiplier = self._get_rescale_multiplier_for_key(key)
484
                s *= multiplier
485
486
                samples[:, idx] = s

487
488
        return samples

489
    def _get_labels(self):
490
        """ Combine the units, symbols and rescaling to give labels """
491

492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
        labels = []
        for key in self.theta_keys:
            label = None
            s = self.symbol_dictionary[key]
            s.replace('_{glitch}', r'_\textrm{glitch}')
            u = self.unit_dictionary[key]
            if key in self.rescale_dictionary:
                if 'symbol' in self.rescale_dictionary[key]:
                    s = self.rescale_dictionary[key]['symbol']
                if 'label' in self.rescale_dictionary[key]:
                    label = self.rescale_dictionary[key]['label']
                if 'unit' in self.rescale_dictionary[key]:
                    u = self.rescale_dictionary[key]['unit']
            if label is None:
                label = '{} \n [{}]'.format(s, u)
            labels.append(label)
        return labels
509

510
511
    def plot_corner(self, figsize=(7, 7), add_prior=False, nstds=None,
                    label_offset=0.4, dpi=300, rc_context={},
512
                    tglitch_ratio=False, fig_and_axes=None, save_fig=True,
513
                    **kwargs):
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
        """ Generate a corner plot of the posterior

        Using the `corner` package (https://pypi.python.org/pypi/corner/),
        generate estimates of the posterior from the production samples.

        Parameters
        ----------
        figsize: tuple (7, 7)
            Figure size in inches (passed to plt.subplots)
        add_prior: bool
            If true, plot the prior as a red line
        nstds: float
            The number of standard deviations to plot centered on the mean
        label_offset: float
            Offset the labels from the plot: useful to precent overlapping the
            tick labels with the axis labels
        dpi: int
            Passed to plt.savefig
        rc_context: dict
            Dictionary of rc values to set while generating the figure (see
            matplotlib rc for more details)
        tglitch_ratio: bool
            If true, and tglitch is a parameter, plot posteriors as the
            fractional time at which the glitch occurs instead of the actual
            time
539
540
541
542
543
        fig_and_axes: tuple
            fig and axes to plot on, the axes must be of the right shape,
            namely (ndim, ndim)
        save_fig: bool
            If true, save the figure, else return the fig, axes
544
545
546
547

        Note: kwargs are passed on to corner.coner

        """
548

549
550
551
552
        if 'truths' in kwargs and len(kwargs['truths']) != self.ndim:
            logging.warning('len(Truths) != ndim, Truths will be ignored')
            kwargs['truths'] = None

Gregory Ashton's avatar
Gregory Ashton committed
553
554
        if self.ndim < 2:
            with plt.rc_context(rc_context):
555
556
557
558
                if fig_and_axes is None:
                    fig, ax = plt.subplots(figsize=figsize)
                else:
                    fig, ax = fig_and_axes
Gregory Ashton's avatar
Gregory Ashton committed
559
560
561
562
563
564
565
                ax.hist(self.samples, bins=50, histtype='stepfilled')
                ax.set_xlabel(self.theta_symbols[0])

            fig.savefig('{}/{}_corner.png'.format(
                self.outdir, self.label), dpi=dpi)
            return

566
        with plt.rc_context(rc_context):
567
568
569
570
571
            if fig_and_axes is None:
                fig, axes = plt.subplots(self.ndim, self.ndim,
                                         figsize=figsize)
            else:
                fig, axes = fig_and_axes
572
573

            samples_plt = copy.copy(self.samples)
574
            labels = self._get_labels()
575

576
            samples_plt = self._scale_samples(samples_plt, self.theta_keys)
577
578
579
580
581

            if tglitch_ratio:
                for j, k in enumerate(self.theta_keys):
                    if k == 'tglitch':
                        s = samples_plt[:, j]
582
583
584
                        samples_plt[:, j] = (
                            s - self.minStartTime)/(
                                self.maxStartTime - self.minStartTime)
585
                        labels[j] = r'$R_{\textrm{glitch}}$'
586
587
588
589
590
591
592

            if type(nstds) is int and 'range' not in kwargs:
                _range = []
                for j, s in enumerate(samples_plt.T):
                    median = np.median(s)
                    std = np.std(s)
                    _range.append((median - nstds*std, median + nstds*std))
593
594
            elif 'range' in kwargs:
                _range = kwargs.pop('range')
595
596
597
598
            else:
                _range = None

            fig_triangle = corner.corner(samples_plt,
599
                                         labels=labels,
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
                                         fig=fig,
                                         bins=50,
                                         max_n_ticks=4,
                                         plot_contours=True,
                                         plot_datapoints=True,
                                         label_kwargs={'fontsize': 8},
                                         data_kwargs={'alpha': 0.1,
                                                      'ms': 0.5},
                                         range=_range,
                                         **kwargs)

            axes_list = fig_triangle.get_axes()
            axes = np.array(axes_list).reshape(self.ndim, self.ndim)
            plt.draw()
            for ax in axes[:, 0]:
                ax.yaxis.set_label_coords(-label_offset, 0.5)
            for ax in axes[-1, :]:
                ax.xaxis.set_label_coords(0.5, -label_offset)
            for ax in axes_list:
                ax.set_rasterized(True)
                ax.set_rasterization_zorder(-10)
            plt.tight_layout(h_pad=0.0, w_pad=0.0)
            fig.subplots_adjust(hspace=0.05, wspace=0.05)

            if add_prior:
625
                self._add_prior_to_corner(axes, self.samples)
626

627
628
629
630
631
            if save_fig:
                fig_triangle.savefig('{}/{}_corner.png'.format(
                    self.outdir, self.label), dpi=dpi)
            else:
                return fig, axes
632

633
    def _add_prior_to_corner(self, axes, samples):
634
635
636
637
        for i, key in enumerate(self.theta_keys):
            ax = axes[i][i]
            xlim = ax.get_xlim()
            s = samples[:, i]
638
            prior = self._generic_lnprior(**self.theta_prior[key])
639
            x = np.linspace(s.min(), s.max(), 100)
640
641
            multiplier = self._get_rescale_multiplier_for_key(key)
            subtractor = self._get_rescale_subtractor_for_key(key)
642
643
            ax2 = ax.twinx()
            ax2.get_yaxis().set_visible(False)
644
645
            ax2.plot((x-subtractor)*multiplier, [prior(xi) for xi in x], '-r')
            ax2.set_xlim(xlim)
646

647
648
649
650
651
652
653
654
    def plot_prior_posterior(self, normal_stds=2):
        """ Plot the posterior in the context of the prior """
        fig, axes = plt.subplots(nrows=self.ndim, figsize=(8, 4*self.ndim))
        N = 1000
        from scipy.stats import gaussian_kde

        for i, (ax, key) in enumerate(zip(axes, self.theta_keys)):
            prior_dict = self.theta_prior[key]
655
            prior_func = self._generic_lnprior(**prior_dict)
656
657
658
659
660
661
662
663
664
665
            if prior_dict['type'] == 'unif':
                x = np.linspace(prior_dict['lower'], prior_dict['upper'], N)
                prior = prior_func(x)
                prior[0] = 0
                prior[-1] = 0
            elif prior_dict['type'] == 'norm':
                lower = prior_dict['loc'] - normal_stds * prior_dict['scale']
                upper = prior_dict['loc'] + normal_stds * prior_dict['scale']
                x = np.linspace(lower, upper, N)
                prior = prior_func(x)
666
667
668
669
670
            elif prior_dict['type'] == 'halfnorm':
                lower = prior_dict['loc']
                upper = prior_dict['loc'] + normal_stds * prior_dict['scale']
                x = np.linspace(lower, upper, N)
                prior = [prior_func(xi) for xi in x]
Gregory Ashton's avatar
Gregory Ashton committed
671
672
673
674
675
            elif prior_dict['type'] == 'neghalfnorm':
                upper = prior_dict['loc']
                lower = prior_dict['loc'] - normal_stds * prior_dict['scale']
                x = np.linspace(lower, upper, N)
                prior = [prior_func(xi) for xi in x]
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
            else:
                raise ValueError('Not implemented for prior type {}'.format(
                    prior_dict['type']))
            priorln = ax.plot(x, prior, 'r', label='prior')
            ax.set_xlabel(self.theta_symbols[i])

            s = self.samples[:, i]
            while len(s) > 10**4:
                # random downsample to avoid slow calculation of kde
                s = np.random.choice(s, size=int(len(s)/2.))
            kde = gaussian_kde(s)
            ax2 = ax.twinx()
            postln = ax2.plot(x, kde.pdf(x), 'k', label='posterior')
            ax2.set_yticklabels([])
            ax.set_yticklabels([])

        lns = priorln + postln
        labs = [l.get_label() for l in lns]
        axes[0].legend(lns, labs, loc=1, framealpha=0.8)

        fig.savefig('{}/{}_prior_posterior.png'.format(
            self.outdir, self.label))

699
    def plot_cumulative_max(self, **kwargs):
700
701
702
703
        """ Plot the cumulative twoF for the maximum posterior estimate

        See the pyfstat.core.plot_twoF_cumulative function for further details
        """
Gregory Ashton's avatar
Gregory Ashton committed
704
705
706
707
        d, maxtwoF = self.get_max_twoF()
        for key, val in self.theta_prior.iteritems():
            if key not in d:
                d[key] = val
708
709

        if hasattr(self, 'search') is False:
710
            self._initiate_search_object()
711
712
713
        if self.binary is False:
            self.search.plot_twoF_cumulative(
                self.label, self.outdir, F0=d['F0'], F1=d['F1'], F2=d['F2'],
714
                Alpha=d['Alpha'], Delta=d['Delta'],
715
                tstart=self.minStartTime, tend=self.maxStartTime,
716
                **kwargs)
717
718
719
720
721
        else:
            self.search.plot_twoF_cumulative(
                self.label, self.outdir, F0=d['F0'], F1=d['F1'], F2=d['F2'],
                Alpha=d['Alpha'], Delta=d['Delta'], asini=d['asini'],
                period=d['period'], ecc=d['ecc'], argp=d['argp'], tp=d['argp'],
722
                tstart=self.minStartTime, tend=self.maxStartTime, **kwargs)
Gregory Ashton's avatar
Gregory Ashton committed
723

724
    def _generic_lnprior(self, **kwargs):
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
        """ Return a lambda function of the pdf

        Parameters
        ----------
        kwargs: dict
            A dictionary containing 'type' of pdf and shape parameters

        """

        def logunif(x, a, b):
            above = x < b
            below = x > a
            if type(above) is not np.ndarray:
                if above and below:
                    return -np.log(b-a)
                else:
                    return -np.inf
            else:
                idxs = np.array([all(tup) for tup in zip(above, below)])
                p = np.zeros(len(x)) - np.inf
                p[idxs] = -np.log(b-a)
                return p

        def halfnorm(x, loc, scale):
749
            if x < loc:
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
                return -np.inf
            else:
                return -0.5*((x-loc)**2/scale**2+np.log(0.5*np.pi*scale**2))

        def cauchy(x, x0, gamma):
            return 1.0/(np.pi*gamma*(1+((x-x0)/gamma)**2))

        def exp(x, x0, gamma):
            if x > x0:
                return np.log(gamma) - gamma*(x - x0)
            else:
                return -np.inf

        if kwargs['type'] == 'unif':
            return lambda x: logunif(x, kwargs['lower'], kwargs['upper'])
        elif kwargs['type'] == 'halfnorm':
            return lambda x: halfnorm(x, kwargs['loc'], kwargs['scale'])
767
768
        elif kwargs['type'] == 'neghalfnorm':
            return lambda x: halfnorm(-x, kwargs['loc'], kwargs['scale'])
769
770
771
772
773
774
775
        elif kwargs['type'] == 'norm':
            return lambda x: -0.5*((x - kwargs['loc'])**2/kwargs['scale']**2
                                   + np.log(2*np.pi*kwargs['scale']**2))
        else:
            logging.info("kwargs:", kwargs)
            raise ValueError("Print unrecognise distribution")

776
    def _generate_rv(self, **kwargs):
777
778
779
780
781
782
783
784
        dist_type = kwargs.pop('type')
        if dist_type == "unif":
            return np.random.uniform(low=kwargs['lower'], high=kwargs['upper'])
        if dist_type == "norm":
            return np.random.normal(loc=kwargs['loc'], scale=kwargs['scale'])
        if dist_type == "halfnorm":
            return np.abs(np.random.normal(loc=kwargs['loc'],
                                           scale=kwargs['scale']))
785
786
787
        if dist_type == "neghalfnorm":
            return -1 * np.abs(np.random.normal(loc=kwargs['loc'],
                                                scale=kwargs['scale']))
788
789
790
791
792
793
        if dist_type == "lognorm":
            return np.random.lognormal(
                mean=kwargs['loc'], sigma=kwargs['scale'])
        else:
            raise ValueError("dist_type {} unknown".format(dist_type))

794
795
796
    def _plot_walkers(self, sampler, symbols=None, alpha=0.4, color="k",
                      temp=0, lw=0.1, nprod=0, add_det_stat_burnin=False,
                      fig=None, axes=None, xoffset=0, plot_det_stat=False,
797
                      context='ggplot', subtractions=None, labelpad=0.05):
798
799
        """ Plot all the chains from a sampler """

800
801
802
803
804
        if context not in plt.style.available:
            raise ValueError((
                'The requested context {} is not available; please select a'
                ' context from `plt.style.available`').format(context))

805
806
807
        if np.ndim(axes) > 1:
            axes = axes.flatten()

808
809
810
811
812
813
814
815
816
817
818
819
820
        shape = sampler.chain.shape
        if len(shape) == 3:
            nwalkers, nsteps, ndim = shape
            chain = sampler.chain[:, :, :]
        if len(shape) == 4:
            ntemps, nwalkers, nsteps, ndim = shape
            if temp < ntemps:
                logging.info("Plotting temperature {} chains".format(temp))
            else:
                raise ValueError(("Requested temperature {} outside of"
                                  "available range").format(temp))
            chain = sampler.chain[temp, :, :, :]

821
822
        if subtractions is None:
            subtractions = [0 for i in range(ndim)]
823
824
825
        else:
            if len(subtractions) != self.ndim:
                raise ValueError('subtractions must be of length ndim')
826

827
828
829
830
        if plot_det_stat:
            extra_subplots = 1
        else:
            extra_subplots = 0
831
        with plt.style.context((context)):
Gregory Ashton's avatar
Gregory Ashton committed
832
            plt.rcParams['text.usetex'] = True
Gregory Ashton's avatar
Gregory Ashton committed
833
            if fig is None and axes is None:
834
                fig = plt.figure(figsize=(4, 3.0*ndim))
835
836
                ax = fig.add_subplot(ndim+extra_subplots, 1, 1)
                axes = [ax] + [fig.add_subplot(ndim+extra_subplots, 1, i)
Gregory Ashton's avatar
Gregory Ashton committed
837
                               for i in range(2, ndim+1)]
838

Gregory Ashton's avatar
Gregory Ashton committed
839
            idxs = np.arange(chain.shape[1])
840
841
842
843
844
            burnin_idx = chain.shape[1] - nprod
            if hasattr(self, 'convergence_idx'):
                convergence_idx = self.convergence_idx
            else:
                convergence_idx = burnin_idx
845
846
            if ndim > 1:
                for i in range(ndim):
847
                    axes[i].ticklabel_format(useOffset=False, axis='y')
Gregory Ashton's avatar
Gregory Ashton committed
848
                    cs = chain[:, :, i].T
849
                    if burnin_idx > 0:
850
851
                        axes[i].plot(xoffset+idxs[:convergence_idx+1],
                                     cs[:convergence_idx+1]-subtractions[i],
852
                                     color="r", alpha=alpha,
Gregory Ashton's avatar
Gregory Ashton committed
853
                                     lw=lw)
854
                        axes[i].axvline(xoffset+convergence_idx,
855
                                        color='k', ls='--', lw=0.25)
856
857
                    axes[i].plot(xoffset+idxs[burnin_idx:],
                                 cs[burnin_idx:]-subtractions[i],
Gregory Ashton's avatar
Gregory Ashton committed
858
                                 color="k", alpha=alpha, lw=lw)
859
                    if symbols:
860
                        if subtractions[i] == 0:
861
                            axes[i].set_ylabel(symbols[i], labelpad=labelpad)
862
863
                        else:
                            axes[i].set_ylabel(
864
865
                                symbols[i]+'$-$'+symbols[i]+'$_0$',
                                labelpad=labelpad)
866

867
868
                    if hasattr(self, 'convergence_diagnostic'):
                        ax = axes[i].twinx()
869
870
                        c_x = np.array(self.convergence_diagnosticx)
                        c_y = np.array(self.convergence_diagnostic)
871
872
873
874
                        break_idx = np.argmin(np.abs(c_x - burnin_idx))
                        ax.plot(c_x[:break_idx], c_y[:break_idx, i], '-b')
                        ax.plot(c_x[break_idx:], c_y[break_idx:, i], '-b')
                        ax.set_ylabel('PSRF')
875
                        ax.ticklabel_format(useOffset=False)
876
                        ax.set_ylim(1, self.convergence_plot_upper_lim)
877
            else:
Gregory Ashton's avatar
Gregory Ashton committed
878
                axes[0].ticklabel_format(useOffset=False, axis='y')
Gregory Ashton's avatar
Gregory Ashton committed
879
                cs = chain[:, :, temp].T
Gregory Ashton's avatar
Gregory Ashton committed
880
881
882
883
884
885
                if burnin_idx:
                    axes[0].plot(idxs[:burnin_idx], cs[:burnin_idx],
                                 color="r", alpha=alpha, lw=lw)
                axes[0].plot(idxs[burnin_idx:], cs[burnin_idx:], color="k",
                             alpha=alpha, lw=lw)
                if symbols:
886
                    axes[0].set_ylabel(symbols[0], labelpad=labelpad)
887

888
            if plot_det_stat:
889
890
891
                if len(axes) == ndim:
                    axes.append(fig.add_subplot(ndim+1, 1, ndim+1))

892
893
894
                lnl = sampler.lnlikelihood[temp, :, :]
                if burnin_idx and add_det_stat_burnin:
                    burn_in_vals = lnl[:, :burnin_idx].flatten()
895
896
897
898
899
900
901
                    try:
                        axes[-1].hist(burn_in_vals[~np.isnan(burn_in_vals)],
                                      bins=50, histtype='step', color='r')
                    except ValueError:
                        logging.info('Det. Stat. hist failed, most likely all '
                                     'values where the same')
                        pass
902
903
904
                else:
                    burn_in_vals = []
                prod_vals = lnl[:, burnin_idx:].flatten()
905
906
907
908
909
910
911
                try:
                    axes[-1].hist(prod_vals[~np.isnan(prod_vals)], bins=50,
                                  histtype='step', color='k')
                except ValueError:
                    logging.info('Det. Stat. hist failed, most likely all '
                                 'values where the same')
                    pass
912
913
914
915
916
917
918
919
920
921
922
923
                if self.BSGL:
                    axes[-1].set_xlabel(r'$\mathcal{B}_\mathrm{S/GL}$')
                else:
                    axes[-1].set_xlabel(r'$\widetilde{2\mathcal{F}}$')
                axes[-1].set_ylabel(r'$\textrm{Counts}$')
                combined_vals = np.append(burn_in_vals, prod_vals)
                if len(combined_vals) > 0:
                    minv = np.min(combined_vals)
                    maxv = np.max(combined_vals)
                    Range = abs(maxv-minv)
                    axes[-1].set_xlim(minv-0.1*Range, maxv+0.1*Range)

924
                xfmt = matplotlib.ticker.ScalarFormatter()
925
                xfmt.set_powerlimits((-4, 4))
926
927
                axes[-1].xaxis.set_major_formatter(xfmt)

928
            axes[-2].set_xlabel(r'$\textrm{Number of steps}$', labelpad=0.2)
929
930
        return fig, axes

931
    def _apply_corrections_to_p0(self, p0):
Gregory Ashton's avatar
Gregory Ashton committed
932
933
934
        """ Apply any correction to the initial p0 values """
        return p0

935
    def _generate_scattered_p0(self, p):
936
        """ Generate a set of p0s scattered about p """
Gregory Ashton's avatar
Gregory Ashton committed
937
        p0 = [[p + self.scatter_val * p * np.random.randn(self.ndim)
938
939
940
941
               for i in xrange(self.nwalkers)]
              for j in xrange(self.ntemps)]
        return p0

942
    def _generate_initial_p0(self):
943
944
945
        """ Generate a set of init vals for the walkers """

        if type(self.theta_initial) == dict:
946
            logging.info('Generate initial values from initial dictionary')
947
            if hasattr(self, 'nglitch') and self.nglitch > 1:
948
                raise ValueError('Initial dict not implemented for nglitch>1')
949
            p0 = [[[self._generate_rv(**self.theta_initial[key])
950
951
952
                    for key in self.theta_keys]
                   for i in range(self.nwalkers)]
                  for j in range(self.ntemps)]
953
954
        elif type(self.theta_initial) == list:
            logging.info('Generate initial values from list of theta_initial')
955
            p0 = [[[self._generate_rv(**val)
956
957
958
                    for val in self.theta_initial]
                   for i in range(self.nwalkers)]
                  for j in range(self.ntemps)]
959
        elif self.theta_initial is None:
960
            logging.info('Generate initial values from prior dictionary')
961
            p0 = [[[self._generate_rv(**self.theta_prior[key])
962
963
964
965
                    for key in self.theta_keys]
                   for i in range(self.nwalkers)]
                  for j in range(self.ntemps)]
        elif len(self.theta_initial) == self.ndim:
966
            p0 = self._generate_scattered_p0(self.theta_initial)
967
968
969
970
971
        else:
            raise ValueError('theta_initial not understood')

        return p0

972
    def _get_new_p0(self, sampler):
973
974
975
976
977
978
        """ Returns new initial positions for walkers are burn0 stage

        This returns new positions for all walkers by scattering points about
        the maximum posterior with scale `scatter_val`.

        """
Gregory Ashton's avatar
Gregory Ashton committed
979
980
981
982
        temp_idx = 0
        pF = sampler.chain[temp_idx, :, :, :]
        lnl = sampler.lnlikelihood[temp_idx, :, :]
        lnp = sampler.lnprobability[temp_idx, :, :]
983
984

        # General warnings about the state of lnp
Gregory Ashton's avatar
Gregory Ashton committed
985
        if np.any(np.isnan(lnp)):
986
987
            logging.warning(
                "Of {} lnprobs {} are nan".format(
Gregory Ashton's avatar
Gregory Ashton committed
988
989
                    np.shape(lnp), np.sum(np.isnan(lnp))))
        if np.any(np.isposinf(lnp)):
990
991
            logging.warning(
                "Of {} lnprobs {} are +np.inf".format(
Gregory Ashton's avatar
Gregory Ashton committed
992
993
                    np.shape(lnp), np.sum(np.isposinf(lnp))))
        if np.any(np.isneginf(lnp)):
994
995
            logging.warning(
                "Of {} lnprobs {} are -np.inf".format(
Gregory Ashton's avatar
Gregory Ashton committed
996
                    np.shape(lnp), np.sum(np.isneginf(lnp))))
997

998
999
        lnp_finite = copy.copy(lnp)
        lnp_finite[np.isinf(lnp)] = np.nan
Gregory Ashton's avatar
Gregory Ashton committed
1000
1001
        idx = np.unravel_index(np.nanargmax(lnp_finite), lnp_finite.shape)
        p = pF[idx]
1002
        p0 = self._generate_scattered_p0(p)
1003

1004
1005
1006
1007
1008
1009
1010
1011
        self.search.BSGL = False
        twoF = self.logl(p, self.search)
        self.search.BSGL = self.BSGL

        logging.info(('Gen. new p0 from pos {} which had det. stat.={:2.1f},'
                      ' twoF={:2.1f} and lnp={:2.1f}')
                     .format(idx[1], lnl[idx], twoF, lnp_finite[idx]))

1012
1013
        return p0

1014
    def _get_data_dictionary_to_save(self):
1015
1016
        d = dict(nsteps=self.nsteps, nwalkers=self.nwalkers,
                 ntemps=self.ntemps, theta_keys=self.theta_keys,
Gregory Ashton's avatar
Gregory Ashton committed
1017
                 theta_prior=self.theta_prior, scatter_val=self.scatter_val,
1018
                 log10temperature_min=self.log10temperature_min,
1019
                 BSGL=self.BSGL)
1020
1021
        return d

1022
    def _save_data(self, sampler, samples, lnprobs, lnlikes, all_lnlikelihood):
1023
        d = self._get_data_dictionary_to_save()
1024
1025
1026
        d['samples'] = samples
        d['lnprobs'] = lnprobs
        d['lnlikes'] = lnlikes
1027
        d['all_lnlikelihood'] = all_lnlikelihood
1028
1029
1030
1031
1032
1033
1034
1035

        if os.path.isfile(self.pickle_path):
            logging.info('Saving backup of {} as {}.old'.format(
                self.pickle_path, self.pickle_path))
            os.rename(self.pickle_path, self.pickle_path+".old")
        with open(self.pickle_path, "wb") as File:
            pickle.dump(d, File)

1036
1037
    def get_saved_data_dictionary(self):
        """ Returns dictionary of the data saved as pickle """
1038
1039
1040
1041
        with open(self.pickle_path, "r") as File:
            d = pickle.load(File)
        return d

1042
    def _check_old_data_is_okay_to_use(self):
1043
1044
1045
1046
        if args.use_old_data:
            logging.info("Forcing use of old data")
            return True

1047
1048
1049
1050
        if os.path.isfile(self.pickle_path) is False:
            logging.info('No pickled data found')
            return False

Gregory Ashton's avatar
Gregory Ashton committed
1051
1052
        if self.sftfilepath is not None:
            oldest_sft = min([os.path.getmtime(f) for f in
1053
                              self._get_list_of_matching_sfts()])
Gregory Ashton's avatar
Gregory Ashton committed
1054
1055
1056
            if os.path.getmtime(self.pickle_path) < oldest_sft:
                logging.info('Pickled data outdates sft files')
                return False
1057

1058
1059
        old_d = self.get_saved_data_dictionary().copy()
        new_d = self._get_data_dictionary_to_save().copy()
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070

        old_d.pop('samples')
        old_d.pop('lnprobs')
        old_d.pop('lnlikes')

        mod_keys = []
        for key in new_d.keys():
            if key in old_d:
                if new_d[key] != old_d[key]:
                    mod_keys.append((key, old_d[key], new_d[key]))
            else:
1071
                raise ValueError('Keys {} not in old dictionary'.format(key))
1072
1073
1074
1075
1076
1077
1078
1079
1080

        if len(mod_keys) == 0:
            return True
        else:
            logging.warning("Saved data differs from requested")
            logging.info("Differences found in following keys:")
            for key in mod_keys:
                if len(key) == 3:
                    if np.isscalar(key[1]) or key[0] == 'nsteps':
1081
                        logging.info("    {} : {} -> {}".format(*key))
1082
                    else:
1083
                        logging.info("    " + key[0])
1084
1085
1086
1087
1088
                else:
                    logging.info(key)
            return False<