def test_parallel_apply():
array = numpy.linspace(-1, 1, 100)
with pytest.raises(TypeError):
parallel_apply(f, array, wrong_argument=None)
with pytest.raises(ValueError):
parallel_apply(f, array, parallel='wrong_argument')
for par in [False, True, -1, 2]:
farray = parallel_apply(f, array, parallel=par)
assert_allclose([f(x) for x in array], farray)
def compute_samples(f, x, samples, **kwargs):
r""" Apply f(x,theta) to x array and theta in samples.
Parameters
----------
f: function
list of functions :math:`f(x;\theta)` with dependent variable
:math:`x`, parameterised by :math:`\theta`.
x: 1D array-like
x values to evaluate :math:`f(x;\theta)` at.
samples: 2D array-like
list of theta samples to evaluate :math:`f(x;\theta)` at.
`shape = (nfunc, nsamples, npars)`
parallel, tqdm_kwargs: optional
see docstring for :func:`fgivenx.parallel.parallel_apply`
cache: str, optional
File root for saving previous calculations for re-use
default None
Returns
-------
2D numpy.array:
samples at each x. `shape=(len(x),len(samples),)`
"""
parallel = kwargs.pop('parallel', False)
cache = kwargs.pop('cache', '')
tqdm_kwargs = kwargs.pop('tqdm_kwargs', {})
if kwargs:
raise TypeError('Unexpected **kwargs: %r' % kwargs)
if cache:
cache = Cache(cache + '_fsamples')
try:
return cache.check(x, samples)
except CacheException as e:
print(e)
fsamples = []
for fi, s in zip(f, samples):
if len(s) > 0:
fsamps = parallel_apply(fi,
s,
precurry=(x, ),
parallel=parallel,
tqdm_kwargs=tqdm_kwargs)
fsamps = numpy.array(fsamps).transpose().copy()
fsamples.append(fsamps)
fsamples = numpy.concatenate(fsamples, axis=1)
if cache:
cache.save(x, samples, fsamples)
return fsamples
def compute_dkl(fsamps, prior_fsamps, **kwargs):
"""
Compute the Kullback Leibler divergence for function samples for posterior
and prior pre-calculated at a range of x values.
Parameters
----------
fsamps: 2D numpy.array
Posterior function samples, as computed by
:func:`fgivenx.compute_samples`
prior_fsamps: 2D numpy.array
Prior function samples, as computed by :func:`fgivenx.compute_samples`
parallel, tqdm_kwargs: optional
see docstring for :func:`fgivenx.parallel.parallel_apply`.
cache: str, optional
File root for saving previous calculations for re-use.
Returns
-------
1D numpy.array:
Kullback-Leibler divergences at each value of x. `shape=(len(fsamps))`
"""
parallel = kwargs.pop('parallel', False)
cache = kwargs.pop('cache', '')
tqdm_kwargs = kwargs.pop('tqdm_kwargs', {})
if kwargs:
raise TypeError('Unexpected **kwargs: %r' % kwargs)
if cache:
cache = Cache(cache + '_dkl')
try:
return cache.check(fsamps, prior_fsamps)
except CacheException as e:
print(e)
zip_fsamps = list(zip(fsamps, prior_fsamps))
dkls = parallel_apply(DKL,
zip_fsamps,
parallel=parallel,
tqdm_kwargs=tqdm_kwargs)
dkls = numpy.array(dkls)
if cache:
cache.save(fsamps, prior_fsamps, dkls)
return dkls
def compute_pmf(fsamps, y, **kwargs):
""" Compute the pmf defined by fsamps at each x for each y.
Parameters
----------
fsamps: 2D array-like
array of function samples, as returned by
:func:`fgivenx.compute_samples`
y: 1D array-like
y values to evaluate the PMF at
parallel, tqdm_kwargs: optional
see docstring for :func:`fgivenx.parallel.parallel_apply`.
Returns
-------
2D numpy.array
probability mass function at each x for each y
`shape=(len(fsamps),len(y)`
"""
parallel = kwargs.pop('parallel', False)
cache = kwargs.pop('cache', '')
tqdm_kwargs = kwargs.pop('tqdm_kwargs', {})
if kwargs:
raise TypeError('Unexpected **kwargs: %r' % kwargs)
if cache:
cache = Cache(cache + '_masses')
try:
return cache.check(fsamps, y)
except CacheException as e:
print(e)
masses = parallel_apply(PMF,
fsamps,
postcurry=(y, ),
parallel=parallel,
tqdm_kwargs=tqdm_kwargs)
masses = numpy.array(masses).transpose().copy()
if cache:
cache.save(fsamps, y, masses)
return masses