def testLoads(self):
# test initiating from multiple chain arrays
samps = []
for i in range(3):
samps.append(Gaussian2D([1.5, -2], np.diagflat([1, 2])).MCSamples(1001 + i * 10, names=['x', 'y']))
fromChains = MCSamples(samples=[s.samples for s in samps], names=['x', 'y'])
mean = np.sum([s.norm * s.mean('x') for s in samps]) / np.sum([s.norm for s in samps])
meanChains = fromChains.mean('x')
self.assertAlmostEqual(mean, meanChains)
def testLoads(self):
# test initiating from multiple chain arrays
samps = []
for i in range(3):
samps.append(Gaussian2D([1.5, -2], np.diagflat([1, 2])).MCSamples(1001 + i * 10, names=['x', 'y']))
fromChains = MCSamples(samples=[s.samples for s in samps], names=['x', 'y'])
mean = np.sum([s.norm * s.mean('x') for s in samps]) / np.sum([s.norm for s in samps])
meanChains = fromChains.mean('x')
self.assertAlmostEqual(mean, meanChains)
def _init_samples(self,
mcmc_steps=5000,
mcmc_batch_size=5,
ignore_rows=0.3):
u = 2 * (np.random.uniform(size=(mcmc_batch_size, self.x_dim)) - 0.5)
v = self.transform(u)
logl = self.loglike(v)
samples = []
likes = []
for i in range(mcmc_steps):
du = np.random.standard_normal(u.shape) * 0.1
u_prime = u + du
v_prime = self.transform(u_prime)
log_ratio_1 = np.zeros(mcmc_batch_size)
prior = np.logical_or(np.abs(u) > 1, np.abs(u_prime) > 1)
idx = np.where([np.any(p) for p in prior])
log_ratio_1[idx] = -np.inf
rnd_u = np.random.rand(mcmc_batch_size)
ratio = np.clip(np.exp(log_ratio_1), 0, 1)
mask = (rnd_u < ratio).astype(int)
logl_prime = np.full(mcmc_batch_size, logl)
for idx, im in enumerate(mask):
if im:
lp = self.loglike(v_prime[idx])
if lp >= logl[idx]:
logl_prime[idx] = lp
elif rnd_u[idx] < np.clip(np.exp(lp - logl[idx]), 0, 1):
logl_prime[idx] = lp
else:
mask[idx] = 0
m = mask[:, None]
u = u_prime * m + u * (1 - m)
v = v_prime * m + v * (1 - m)
logl = logl_prime * mask + logl * (1 - mask)
samples.append(v)
likes.append(logl)
samples = np.transpose(np.array(samples), axes=[1, 0, 2])
loglikes = -np.transpose(np.array(likes), axes=[1, 0])
weights = np.ones(loglikes.shape)
self._chain_stats(samples)
self._save_samples(samples, weights, loglikes)
names = ['p%i' % i for i in range(int(self.x_dim))]
labels = [r'x_%i' % i for i in range(int(self.x_dim))]
files = chainFiles(os.path.join(self.logs['chains'], 'chain'),
first_chain=1,
last_chain=mcmc_batch_size)
mc = MCSamples(self.logs['chains'],
names=names,
labels=labels,
ignore_rows=ignore_rows)
mc.readChains(files)
return mc
def _read_samples(self, fileroot, match='', ignore_rows=0.3):
names = ['p%i' % i for i in range(int(self.num_params))]
labels = [r'x_%i' % i for i in range(int(self.num_params))]
if match:
files = glob.glob(os.path.join(fileroot, match))
else:
files = chainFiles(fileroot)
mc = MCSamples(fileroot,
names=names,
labels=labels,
ignore_rows=ignore_rows)
mc.readChains(files)
return mc
def MCSamples(self,
size,
names=None,
logLikes=False,
random_state=None,
**kwargs):
"""
Gets a set of independent samples from the mixture as a :class:`.mcsamples.MCSamples` object
ready for plotting etc.
:param size: number of samples
:param names: set to override existing names
:param logLikes: if True set the sample likelihood values from the pdf, if false, don't store log likelihoods
:param random_state: random seed or Generator
:return: a new :class:`.mcsamples.MCSamples` instance
"""
samples = self.sim(size, random_state=random_state)
if logLikes:
loglikes = -np.log(self.pdf(samples))
else:
loglikes = None
return MCSamples(samples=samples,
loglikes=loglikes,
paramNamesFile=copy.deepcopy(self.paramNames),
names=names,
ranges=self.lims,
**kwargs)
def MCSamplesFromCobaya(info, collections, name_tag=None,
ignore_rows=0, ini=None, settings=None):
"""
Creates a set of samples from Cobaya's output.
Parameter names, ranges and labels are taken from the "info" dictionary
(always use the "updated" one generated by `cobaya.run`).
For a description of the various analysis settings and default values see
`analysis_defaults.ini <https://getdist.readthedocs.org/en/latest/analysis_settings.html>`_.
:param collections: collection(s) of samples from Cobaya
:param info: info dictionary, common to all collections
(use the "updated" one, returned by `cobaya.run`)
:param name_tag: name for this sample to be shown in the plots' legend
:param ignore_rows: initial samples to skip, number (`int>=1`) or fraction (`float<1`)
:param ini: The name of a .ini file with analysis settings to use
:param settings: dictionary of analysis settings to override defaults
:return: The :class:`MCSamples` instance
"""
if hasattr(collections, "data"):
collections = [collections]
# Check consistency between collections
try:
columns = list(collections[0].data)
except AttributeError:
raise TypeError(
"The second argument does not appear to be a (list of) samples `Collection`.")
if not all(list(c.data) == columns for c in collections[1:]):
raise ValueError("The given collections don't have the same columns.")
# Check consistency with info
info_params = get_info_params(info)
# if skip burn in *has already been done*
skip = info.get(_post, {}).get("skip", 0)
if ignore_rows != 0 and skip != 0:
logging.warning("You are asking for rows to be ignored (%r), but some (%r) were "
"already ignored in the original chain.", ignore_rows, skip)
var_params = [k for k, v in info_params.items() if is_sampled_param(v) or is_derived_param(v)]
assert set(columns[2:]) == set(var_params), (
"Info and collection(s) are not compatible, because their parameters differ: "
"the collection(s) have %r and the info has %r. " % (columns[2:], var_params) +
"Are you sure that you are using an *updated* info dictionary "
"(i.e. the output of `cobaya.run`)?")
# We need to use *collection* sorting, not info sorting!
names = [p + ("*" if is_derived_param(info_params[p]) else "")
for p in columns[2:]]
labels = [(info_params[p] or {}).get(_p_label, p) for p in columns[2:]]
ranges = {p: get_range(info_params[p]) for p in info_params} # include fixed parameters not in columns
renames = {p: info_params.get(p, {}).get(_p_renames, []) for p in columns[2:]}
samples = [c[c.data.columns[2:]].values for c in collections]
weights = [c[_weight].values for c in collections]
loglikes = [-c[_minuslogpost].values for c in collections]
sampler = get_sampler_type(info)
label = get_sample_label(info)
from getdist.mcsamples import MCSamples
return MCSamples(samples=samples, weights=weights, loglikes=loglikes, sampler=sampler,
names=names, labels=labels, ranges=ranges, renames=renames,
ignore_rows=ignore_rows, name_tag=name_tag, label=label, ini=ini,
settings=settings)
def testDensitySymmetries(self):
# check flipping samples gives flipped density
samps = Gaussian1D(0, 1, xmin=-1, xmax=4).MCSamples(12000)
d = samps.get1DDensity('x')
samps.samples[:, 0] *= -1
samps = MCSamples(samples=samps.samples, names=['x'], ranges={'x': [-4, 1]})
d2 = samps.get1DDensity('x')
self.assertTrue(np.allclose(d.P, d2.P[::-1]))
samps = Gaussian2D([0, 0], np.diagflat([1, 2]), xmin=-1, xmax=2, ymin=0, ymax=3).MCSamples(12000)
d = samps.get2DDensity('x', 'y')
samps.samples[:, 0] *= -1
samps = MCSamples(samples=samps.samples, names=['x', 'y'], ranges={'x': [-2, 1], 'y': [0, 3]})
d2 = samps.get2DDensity('x', 'y')
self.assertTrue(np.allclose(d.P, d2.P[:, ::-1]))
samps.samples[:, 0] *= -1
samps.samples[:, 1] *= -1
samps = MCSamples(samples=samps.samples, names=['x', 'y'], ranges={'x': [-1, 2], 'y': [-3, 0]})
d2 = samps.get2DDensity('x', 'y')
self.assertTrue(np.allclose(d.P, d2.P[::-1, ::], atol=1e-5))
def MCSamples(self, size, names=None, logLikes=False, **kwargs):
"""
Gets a set of independent samples from the mixture as a :class:`.mcsamples.MCSamples` object ready for plotting etc.
:param size: number of samples
:param names: set to override existing names
:param logLikes: if True set the sample likelihood values from the pdf, if false, don't store log likelihoods
:return: list of [x,y] pair names
"""
samples = self.sim(size)
if logLikes:
loglikes = -np.log(self.pdf(samples))
else:
loglikes = None
return MCSamples(samples=samples, loglikes=loglikes, paramNamesFile=self.paramNames, names=names,
ranges=self.lims, **kwargs)
def testDensitySymmetries(self):
# check flipping samples gives flipped density
samps = Gaussian1D(0, 1, xmin=-1, xmax=4).MCSamples(12000)
d = samps.get1DDensity('x')
samps.samples[:, 0] *= -1
samps = MCSamples(samples=samps.samples, names=['x'], ranges={'x': [-4, 1]})
d2 = samps.get1DDensity('x')
self.assertTrue(np.allclose(d.P, d2.P[::-1]))
samps = Gaussian2D([0, 0], np.diagflat([1, 2]), xmin=-1, xmax=2, ymin=0, ymax=3).MCSamples(12000)
d = samps.get2DDensity('x', 'y')
samps.samples[:, 0] *= -1
samps = MCSamples(samples=samps.samples, names=['x', 'y'], ranges={'x': [-2, 1], 'y': [0, 3]})
d2 = samps.get2DDensity('x', 'y')
self.assertTrue(np.allclose(d.P, d2.P[:, ::-1]))
samps.samples[:, 0] *= -1
samps.samples[:, 1] *= -1
samps = MCSamples(samples=samps.samples, names=['x', 'y'], ranges={'x': [-1, 2], 'y': [-3, 0]})
d2 = samps.get2DDensity('x', 'y')
self.assertTrue(np.allclose(d.P, d2.P[::-1, ::], atol=1e-5))
def run(self,
train_iters=200,
mcmc_steps=5000,
bootstrap_iters=1,
bootstrap_mcmc_steps=5000,
bootstrap_fileroot='',
bootstrap_match='',
bootstrap_batch_size=5,
alpha=0,
single_thin=1,
ignore_rows=0.3):
if alpha == 0.0:
alpha = 1 / self.x_dim**0.5
if self.log:
self.logger.info('Alpha [%5.4f]' % (alpha))
for t in range(bootstrap_iters):
if t == 0:
if bootstrap_fileroot:
mc = self._read_samples(bootstrap_fileroot,
match=bootstrap_match,
ignore_rows=ignore_rows)
else:
mc = self._init_samples(
mcmc_steps=bootstrap_mcmc_steps,
mcmc_batch_size=bootstrap_batch_size,
ignore_rows=ignore_rows)
else:
samples, likes, scale, nc = self.trainer.sample(
loglike=self.loglike,
transform=transform,
mcmc_steps=bootstrap_mcmc_steps,
alpha=alpha,
dynamic=False,
show_progress=True)
samples = transform(samples)
self._chain_stats(samples)
loglikes = -np.array(likes)
weights = np.ones(loglikes.shape)
mc = MCSamples(samples=[samples[0]],
weights=[weights[0]],
loglikes=[loglikes[0]],
ignore_rows=ignore_rows)
samples = mc.makeSingleSamples(single_thin=single_thin)
samples = samples[:, :self.x_dim]
mean = np.mean(samples, axis=0)
std = np.std(samples, axis=0)
samples = (samples - mean) / std
self.trainer.train(samples, max_iters=train_iters, noise=-1)
def transform(x):
return x * std + mean
samples, likes, scale, nc = self.trainer.sample(
loglike=self.loglike,
transform=transform,
mcmc_steps=mcmc_steps,
alpha=alpha,
dynamic=False,
show_progress=True,
out_chain=os.path.join(self.logs['chains'], 'chain'))
samples = transform(samples)
self._chain_stats(samples)
def MCSamplesFromCosmosis(chain_root,
chain_min_root=None,
param_name_dict=None,
param_label_dict=None,
name_tag=None,
settings=None):
"""
Function to import Cosmosis chains in GetDist.
:param chain_root: the name and path to the chain or the path to the
folder that contains it.
:param chain_min_root: (optional) name of the file containing the
explicit best fit.
:param param_name_dict: (optional) a dictionary with the mapping between
cosmosis names and reasonable parameter names.
:param param_label_dict: (optional) dictionary with the mapping between
parameter names and parameter labels, since Cosmosis does not save
the labels in the chain.
:param name_tag: (optional) a string with the name tag for the chain.
:param settings: (optional) dictionary of analysis settings to override
getdist defaults
:return: The :class:`~getdist.mcsamples.MCSamples` instance
"""
# decide if the user passed a folder or a chain:
if os.path.isfile(chain_root + '.txt'):
chain_file = chain_root + '.txt'
elif os.path.isdir(chain_root):
# look for the chain file:
temp = list(filter(lambda x: 'chain.txt' in x, os.listdir(chain_root)))
if len(temp) == 0:
raise ValueError('No chain file found in folder', chain_root)
chain_file = chain_root + '/' + temp[0]
else:
raise ValueError('Input chain root is not a folder nor a file.')
# get all the commented lines in the chain file:
info = get_cosmosis_info(chain_file)
# get the parameter names:
param_names = get_param_names(info)
# get the parameter labels from the user provided dictionary:
param_labels = get_param_labels(info, param_names, param_label_dict)
# get the sampler:
sampler = get_sampler_type(info)
# get the name tag:
if name_tag is None:
name_tag = get_name_tag(info)
# get the samples weights and likelihood:
chain = loadNumpyTxt(chain_file, skiprows=0)
# parse the chain depending on the sampler that produced it:
if sampler == 'nested':
# get number of samples to use:
nsamples = int(
list(filter(lambda x: 'nsample=' in x,
info))[0].replace(' ', '').split('=')[1])
# get the chain:
chain = chain[-nsamples:]
# get all quantities:
indexes = [
i for i in range(len(param_names)) if
i != param_names.index('weight') and i != param_names.index('post')
]
samples = chain[:, indexes]
weights = chain[:, param_names.index('weight')]
loglike = chain[:, param_names.index('post')]
# delete the weights and likelihood from names:
if param_labels is not None:
param_labels.pop(param_names.index('weight'))
param_labels.pop(param_names.index('post'))
param_names.pop(param_names.index('weight'))
param_names.pop(param_names.index('post'))
elif sampler == 'mcmc':
# get all quantities:
indexes = [
i for i in range(len(param_names))
if i != param_names.index('post')
]
samples = chain[:, indexes]
loglike = chain[:, param_names.index('post')]
# Cosmosis does not weight samples:
samples, idx, weights = np.unique(samples,
return_index=True,
return_counts=True,
axis=0)
loglike = loglike[idx]
# delete the weights and likelihood from names:
if param_labels is not None:
param_labels.pop(param_names.index('post'))
param_names.pop(param_names.index('post'))
elif sampler == 'uncorrelated':
# get all quantities:
indexes = [
i for i in range(len(param_names))
if i != param_names.index('post')
]
samples = chain[:, indexes]
loglike = chain[:, param_names.index('post')]
weights = None
# delete the weights and likelihood from names:
if param_labels is not None:
param_labels.pop(param_names.index('post'))
param_names.pop(param_names.index('post'))
else:
raise ValueError('Unknown sampler')
# get the ranges:
ranges = get_ranges(info, param_names)
# transform param names:
if param_name_dict is not None:
for i, name in enumerate(param_names):
if name in param_name_dict.keys():
param_names[i] = param_name_dict[name]
if name in ranges.keys():
ranges[param_name_dict[name]] = ranges.pop(name)
#for i, name in enumerate(param_names):
# if name in param_name_dict.keys():
# initialize the samples:
mc_samples = MCSamples(samples=samples,
weights=weights,
loglikes=-2. * loglike,
sampler=sampler,
names=param_names,
labels=param_labels,
ranges=ranges,
ignore_rows=0,
name_tag=name_tag,
settings=settings)
# set running parameters:
for name in mc_samples.getParamNames().parsWithNames(
mc_samples.getParamNames().list()):
if name.name in ranges.keys():
name.isDerived = False
else:
name.isDerived = True
# polish the samples removing nans:
mc_samples = polish_samples(mc_samples)
# get the best fit:
if chain_min_root is not None:
# since getdist does not cache the best fit we have to override the
# method in this brute way:
funcType = types.MethodType
mc_samples.getBestFit = funcType(
functools.partial(get_maximum_likelihood,
chain_min_root=chain_min_root,
param_name_dict=param_name_dict,
param_label_dict=param_label_dict), mc_samples)
# update statistics:
mc_samples.updateBaseStatistics()
#
return mc_samples
def __init__(self,
root,
base_dir,
use_nestcheck,
transform=None,
overwrite_transformed=False,
**kwargs):
filerootpath = _os.path.join(base_dir, root)
_filerootpath = filerootpath
if transform is not None:
samples = _np.loadtxt(filerootpath + '.txt')
ndims = samples.shape[1] - 2
temp = transform(samples[0, 2:], old_API=True)
ntransform = len(temp) - ndims
_exists = _os.path.isfile(filerootpath + '_transformed.txt')
if not _exists or overwrite_transformed:
transformed = _np.zeros(
(samples.shape[0], samples.shape[1] + ntransform))
transformed[:, :2] = samples[:, :2]
for i in range(samples.shape[0]):
transformed[i, 2:] = transform(samples[i, 2:],
old_API=True)
_np.savetxt(filerootpath + '_transformed.txt', transformed)
filerootpath += '_transformed'
root += '_transformed'
super(NestedBackend, self).__init__(filepath=filerootpath + '.txt',
**kwargs)
if getdist is not None:
# getdist backend
self._gd_bcknd = MCSamples(
root=filerootpath,
settings=self.kde_settings,
sampler='nested',
names=self.names,
ranges=self.bounds,
labels=[self.labels[name] for name in self.names])
self._gd_bcknd.readChains(getdist.chains.chainFiles(filerootpath))
self.use_nestcheck = use_nestcheck
if self.use_nestcheck: # nestcheck backend
if transform is not None:
for ext in ['dead-birth.txt', 'phys_live-birth.txt']:
_exists = _os.path.isfile(filerootpath + ext)
if not _exists or overwrite_transformed:
samples = _np.loadtxt(_filerootpath + ext)
transformed = _np.zeros(
(samples.shape[0], samples.shape[1] + ntransform))
transformed[:, ndims + ntransform:] = samples[:,
ndims:]
for i in range(samples.shape[0]):
transformed[i,:ndims+ntransform] =\
transform(samples[i,:ndims],
old_API=True)
_np.savetxt(filerootpath + ext, transformed)
# .stats file with same root needed, but do not need to modify
# the .stats file contents
if not _os.path.isfile(filerootpath + '.stats'):
if _os.path.isfile(_filerootpath + '.stats'):
try:
from shutil import copyfile as _copyfile
except ImportError:
pass
else:
_copyfile(_filerootpath + '.stats',
filerootpath + '.stats')
try:
kwargs['implementation']
except KeyError:
print('Root %r sampling implementation not specified... '
'assuming MultiNest for nestcheck...')
self._nc_bcknd = process_multinest_run(root, base_dir=base_dir)
else:
if kwargs['implementation'] == 'multinest':
self._nc_bcknd = process_multinest_run(root,
base_dir=base_dir)
elif kwargs['implementation'] == 'polychord':
self._nc_bcknd = process_polychord_run(root,
base_dir=base_dir)
else:
raise ValueError('Cannot process with nestcheck.')
class NestedBackend(Run):
"""
Container for nested samples generated by a single run, and backends
for analysis of the run.
The other keyword arguments are generic properties passed to the parent
class, such as the identification (ID) string of the run.
:param str root:
The root filename of the sample file collection.
:param str base_dir:
The directly containing the sample file collection.
:param bool use_nestcheck:
Invoke :mod:`nestcheck` for nested sampling error analysis?
:param callable transform:
A function to transform the parameter vector to another space.
"""
def __init__(self,
root,
base_dir,
use_nestcheck,
transform=None,
overwrite_transformed=False,
**kwargs):
filerootpath = _os.path.join(base_dir, root)
_filerootpath = filerootpath
if transform is not None:
samples = _np.loadtxt(filerootpath + '.txt')
ndims = samples.shape[1] - 2
temp = transform(samples[0, 2:], old_API=True)
ntransform = len(temp) - ndims
_exists = _os.path.isfile(filerootpath + '_transformed.txt')
if not _exists or overwrite_transformed:
transformed = _np.zeros(
(samples.shape[0], samples.shape[1] + ntransform))
transformed[:, :2] = samples[:, :2]
for i in range(samples.shape[0]):
transformed[i, 2:] = transform(samples[i, 2:],
old_API=True)
_np.savetxt(filerootpath + '_transformed.txt', transformed)
filerootpath += '_transformed'
root += '_transformed'
super(NestedBackend, self).__init__(filepath=filerootpath + '.txt',
**kwargs)
if getdist is not None:
# getdist backend
self._gd_bcknd = MCSamples(
root=filerootpath,
settings=self.kde_settings,
sampler='nested',
names=self.names,
ranges=self.bounds,
labels=[self.labels[name] for name in self.names])
self._gd_bcknd.readChains(getdist.chains.chainFiles(filerootpath))
self.use_nestcheck = use_nestcheck
if self.use_nestcheck: # nestcheck backend
if transform is not None:
for ext in ['dead-birth.txt', 'phys_live-birth.txt']:
_exists = _os.path.isfile(filerootpath + ext)
if not _exists or overwrite_transformed:
samples = _np.loadtxt(_filerootpath + ext)
transformed = _np.zeros(
(samples.shape[0], samples.shape[1] + ntransform))
transformed[:, ndims + ntransform:] = samples[:,
ndims:]
for i in range(samples.shape[0]):
transformed[i,:ndims+ntransform] =\
transform(samples[i,:ndims],
old_API=True)
_np.savetxt(filerootpath + ext, transformed)
# .stats file with same root needed, but do not need to modify
# the .stats file contents
if not _os.path.isfile(filerootpath + '.stats'):
if _os.path.isfile(_filerootpath + '.stats'):
try:
from shutil import copyfile as _copyfile
except ImportError:
pass
else:
_copyfile(_filerootpath + '.stats',
filerootpath + '.stats')
try:
kwargs['implementation']
except KeyError:
print('Root %r sampling implementation not specified... '
'assuming MultiNest for nestcheck...')
self._nc_bcknd = process_multinest_run(root, base_dir=base_dir)
else:
if kwargs['implementation'] == 'multinest':
self._nc_bcknd = process_multinest_run(root,
base_dir=base_dir)
elif kwargs['implementation'] == 'polychord':
self._nc_bcknd = process_polychord_run(root,
base_dir=base_dir)
else:
raise ValueError('Cannot process with nestcheck.')
@property
def getdist_backend(self):
""" Get the :class:`getdist.mcsamples.MCSamples` instance. """
return self._gd_bcknd
@property
def nestcheck_backend(self):
""" Get the :mod:`nestcheck` backend for the nested samples. """
return self._nc_bcknd
@property
def margeStats(self):
""" Return the marginal statistics using :mod:`getdist`. """
return self._mcsamples.getMargeStats()
def _add_prior_density(self, plotter, posterior,
ndraws, normalize,
KL_divergence, KL_base,
bootstrap, n_simulate):
""" Crudely estimate the prior density.
Kullback-Leibler divergence estimated in bits for a combined run or
the same run for which the credible intervals are calculated.
"""
run = posterior.subset_to_plot[0]
yield 'Plotting prior for posterior %s...' % posterior.ID
l = posterior.likelihood
if l is None:
return # quietly do nothing
elif not hasattr(l, 'prior'):
return
elif not hasattr(l.prior, 'draw'):
return
elif not callable(l.prior.draw):
return
samples, _ = l.prior.draw(ndraws, transform=True)
color, lw = (run.contours[key] for key in ('color', 'lw'))
quantiles = [None] * 3
with verbose(KL_divergence,
'Estimating 1D marginal KL-divergences in %s' % KL_base,
'Estimated 1D marginal KL-divergences') as condition:
for i, ax in enumerate([plotter.subplots[i,i] \
for i in range(plotter.subplots.shape[0])]):
name = self.params.names[i]
bounds = {name: posterior.bounds[name]}
settings = {'fine_bins': 1024,
'smooth_scale_1D': 0.3,
'boundary_correction_order': 1,
'mult_bias_correction_order': 1} # adopt from posterior settings or take custom input?
idx = l.index(name)
if idx is None: idx = l.prior.index(name)
bcknd = MCSamples(sampler='uncorrelated',
samples=samples[:,idx],
weights=None,
names=[name],
ranges=bounds,
settings=settings)
if normalize:
bcknd.get1DDensity(name).normalize(by='integral',
in_place=True)
x = _np.linspace(ax.xaxis.get_view_interval()[0],
ax.xaxis.get_view_interval()[1],
1000)
ax.plot(x, bcknd.get1DDensity(name).Prob(x),
ls='-.', color=color, lw=lw)
if not condition: continue # go to next iteration if no KL
# a prototype Kullback-Leibler divergence callback
# information in bits
def KL(ns_run, logw):
x = ns_run['theta'][:,posterior.get_index(name)]
w_rel = _np.exp(logw - logw.max())
where = w_rel > run.kde_settings.get('min_weight_ratio',
1.0e-30)
prior = bcknd.get1DDensity(name).Prob(x[where])
p = getdist_kde(x[where], x, w_rel,
ranges=[posterior.bounds[name]],
idx=0,
normalize=normalize,
settings=run.kde_settings)
# Due to spline interpolation, very small densities can be
# negative, so manually give a small postive value which
# does not affect KL integral approximation
p[p<=0.0] = p[p>0.0].min()
KL = _np.sum(w_rel[where] \
* (_np.log(p) - _np.log(prior))) \
/_np.sum(w_rel[where])
if KL_base == 'bits':
return KL / _np.log(2.0)
elif KL_base == 'nats':
return KL
else:
raise ValueError('Invalid base for KL-divergence.')
if bootstrap:
for j, cred_int in enumerate([0.025, 0.5, 0.975]):
quantiles[j] = run_ci_bootstrap(run.nestcheck_backend,
estimator_list=[KL],
cred_int=cred_int,
n_simulate=n_simulate,
simulate_weights=True,
flip_skew=True)
# KL in bits
interval = r'$D_{\mathrm{KL}}=%.2f_{-%.2f}^{+%.2f}$' \
% (quantiles[1],
quantiles[1] - quantiles[0],
quantiles[2] - quantiles[1])
yield ('%s KL-divergence = %.4f/-%.4f/+%.4f'
% (name,
quantiles[1],
quantiles[1] - quantiles[0],
quantiles[2] - quantiles[1]))
if not rcParams['text.usetex']:
fontsize = plotter.settings.lab_fontsize - 1
else:
fontsize = plotter.settings.lab_fontsize
ax.set_title(interval, color=color,
fontsize=fontsize)
else:
where = run.samples[:,0] > 0.0
ns_run = {'theta': run.samples[where,2:]}
divergence = KL(ns_run, _np.log(run.samples[where,0]))
yield ('%s KL-divergence = %.4f' % (name, divergence))
divergence = (r'$D_{\mathrm{KL}}=%.2f$' % divergence)
if not rcParams['text.usetex']:
fontsize = plotter.settings.lab_fontsize - 1
else:
fontsize = plotter.settings.lab_fontsize
ax.set_title(divergence, color=color,
fontsize=fontsize)
yield None
temp = bl[:, :, list(blobs_names).index('temperature')]
print('Last temperature: %s' % temp[-1, -1])
else:
temp = np.ones(ln.shape)
####################################################################################
####################################################################################
# If requested, make getdist-formatted version of the chain
# if args.output_getdist is not None:
if args.output_getdist:
from getdist.mcsamples import MCSamples
gdist = MCSamples(
ranges={par_names[i]: [lbs[i], ubs[i]]
for i in range(n_par)},
samples=np.dstack((ch, bl)).reshape(-1, n_par + n_blobs),
loglikes=(-1. * ln).reshape(-1),
names=list(par_names) + list(blobs_names),
labels=list(par_labels) + list(blobs_labels),
)
gdist.saveAsText(ini_fname_nosuffix + '_gdist')
####################################################################################
####################################################################################
if 1 in plot:
fig1, ax1 = plt.subplots(1, 1)
ax1.plot(-1. * ln, alpha=0.1)
for p, ls in zip([2.5, 16, 50, 84, 97.5], ['-.', '--', '-', '--', '-.']):
ax1.axhline(
np.percentile(-1. * ln, p, axis=1)[n_steps // 2:].mean(),
color='blue',
def bootstrap(
self,
mcmc_steps,
num_walkers,
iters=1,
thin=10,
stats_interval=10,
output_interval=None,
initial_jitter=0.01,
final_jitter=0.01,
init_samples=None,
moves=None):
"""
Args:
num_walkers:
mcmc_steps:
iters:
thin:
stats_interval:
output_interval:
initial_jitter:
final_jitter:
init_samples:
moves:
Returns:
"""
def log_prob(x):
logl, der = self.loglike(x)
return logl + self.prior(x), der
if not os.path.isfile(os.path.join(self.log_dir, 'emcee.h5')) and init_samples is None:
if self.sample_prior is not None:
init_samples = self.sample_prior(num_walkers)
else:
raise ValueError('Prior does not have sample method')
try:
import emcee
except:
raise ImportError
if moves is not None:
ensemble_moves = []
for k, v in moves.items():
if k.lower() == 'stretch':
ensemble_moves.append((emcee.moves.StretchMove(), v))
elif k.lower() == 'kde':
ensemble_moves.append((emcee.moves.KDEMove(), v))
elif k.lower() == 'de':
ensemble_moves.append((emcee.moves.DEMove(), v))
elif k.lower() == 'snooker':
ensemble_moves.append((emcee.moves.DESnookerMove(), v))
else:
ensemble_moves = [(emcee.moves.StretchMove(), 1.0)]
self.logger.info('Performing initial emcee run with [%d] walkers' % (num_walkers))
sampler = emcee.EnsembleSampler(num_walkers, self.x_dim, log_prob, moves=ensemble_moves,
backend=emcee.backends.HDFBackend(os.path.join(self.log_dir, 'emcee.h5')))
state = sampler.run_mcmc(init_samples, mcmc_steps)
self.logger.info('Initial acceptance [%5.4f]' % (np.mean(sampler.acceptance_fraction)))
self._chain_stats(np.transpose(sampler.get_chain(), axes=[1, 0, 2]))
tau = sampler.get_autocorr_time()
training_samples = sampler.get_chain(discard=int(2 * np.max(tau)), flat=True, thin=int(0.5 * np.min(tau)))
for it in range(1, iters + 1):
if iters > 1:
jitter = initial_jitter + (it - 1) * (final_jitter - initial_jitter) / (iters - 1)
else:
jitter = initial_jitter
mean = np.mean(training_samples, axis=0)
std = np.std(training_samples, axis=0)
# Normalise samples
training_samples = (training_samples - mean) / std
self.transform = lambda x: x * std + mean
self.trainer.train(training_samples, jitter=jitter)
init_samples = None
init_loglikes = None
init_derived = None
samples, latent_samples, derived_samples, loglikes, ncall = self._ensemble_sample(
mcmc_steps, num_walkers, init_samples=init_samples,
init_loglikes=init_loglikes, init_derived=init_derived, stats_interval=stats_interval,
output_interval=output_interval)
# Remember last position and loglikes
# init_samples = samples[:, -1, :]
# init_loglikes = loglikes[:, -1]
# init_derived = derived_samples[:, -1, :]
samples = self.transform(samples)
self._chain_stats(samples)
mc = MCSamples(samples=[samples[i, :, :].squeeze() for i in range(samples.shape[0])],
loglikes=[-loglikes[i, :].squeeze() for i in range(loglikes.shape[0])])
training_samples = mc.makeSingleSamples(single_thin=thin)
return training_samples
