def set_dist(self, dist='uniform', kwds=None, dim=None):
r"""
TODO: Add this.
"""
distribution = dist
if (kwds is not None) and (dim is not None):
self.dist.set_dist(distribution, kwds, dim)
elif (kwds is None) and (dim is not None):
self.dist.set_dist(dist=distribution, dim=dim)
# Here we override the default errors printed by scipy.stats with our own.
elif (kwds is None) and (
distributions.supported_distributions(distribution) is 'chi2'):
raise (AttributeError(
"If you are using a chi2 distribution, please pass `df` as a key-value pair in a dictionary. Ex: {'df': 20}."
))
elif (kwds is None) and (
distributions.supported_distributions(distribution) is 'beta'):
raise (AttributeError(
"If you are using a Beta dist, please pass `a` and `b` as key-value pairs in a dictionary. Ex: {'a': 1, 'b': 1}"
))
# the following allows for manual overrides not using the parametric object.
# e.g. kwds = {'loc': [1,2,3]}
elif dim is None:
# print("INPUT: No dimension specified. You will be using `scipy.stats` for your distributions instead of the parametric object. Be warned that functions like `.pdf` may not work as expected.")
if kwds is not None:
self.dist = distributions.assign_dist(distribution, **kwds)
else:
self.dist = distributions.assign_dist(distribution)
pass
def test_supp_dist_norm(self):
print('========== testing distributions.supported_distributions for Normal ==========\n')
assert_equals(type(distributions.supported_distributions('Normal')), _continuous_distns.norm_gen)
assert_equals(type(distributions.supported_distributions('Gauss')), _continuous_distns.norm_gen)
assert_equals(type(distributions.supported_distributions('Gaussian')), _continuous_distns.norm_gen)
assert_equals(type(distributions.supported_distributions('Norm')), _continuous_distns.norm_gen)
assert_equals(type(distributions.supported_distributions('N')), _continuous_distns.norm_gen)
def test_supp_dist_uni(self):
print('========== testing distributions.supported_distributions for Uniform ==========\n')
assert_equals(type(distributions.supported_distributions('Uniform')), _continuous_distns.uniform_gen)
assert_equals(type(distributions.supported_distributions('U')), _continuous_distns.uniform_gen)
assert_equals(type(distributions.supported_distributions('Uni')), _continuous_distns.uniform_gen)
def set_observed_dist(self, distribution=None, kwds=None, dim=None):
r"""
TODO: Add this.
"""
# If `distribution = None`, we query the pushforward density for the top 5% to get a MAP estimate
# TODO print warning about the aforementioned.
# TODO check sizes, ensure dimension agreement
if distribution is not None:
if (kwds is not None) and (dim is not None):
self.observed_dist.set_dist(dim, distribution, kwds)
elif (kwds is None) and (dim is not None):
self.observed_dist.set_dist(dim, distribution)
# Here we override the default errors printed by scipy.stats with our own.
elif (kwds is None) and (
distributions.supported_distributions(distribution) is
'chi2'):
raise (AttributeError(
"If you are using a chi2 distribution, please pass `df` as a key-value pair in a dictionary. Ex: {'df': 20}."
))
elif (kwds is None) and (
distributions.supported_distributions(distribution) is
'beta'):
raise (AttributeError(
"If you are using a Beta dist, please pass `a` and `b` as key-value pairs in a dictionary. Ex: {'a': 1, 'b': 1}"
))
# the following allows for manual overrides not using the parametric object.
# e.g. kwds = {'loc': [1,2,3]}
elif dim is None:
logging.warn(
"OBS: No dimension specified. You will be using `scipy.stats` for your distributions instead of the parametric object. Be warned that functions like `.pdf` may not work as expected."
)
if kwds is not None:
self.observed_dist = distributions.assign_dist(
distribution, **kwds)
else:
self.observed_dist = distributions.assign_dist(
distribution)
else:
logging.warn("""No distribution specified.
Defaulting to normal around data_means with 0.5*data_standard_deviation"""
)
loc = np.mean(self.output.samples, axis=0)
scale = 0.5 * np.std(self.output.samples, axis=0)
self.observed_dist = distributions.assign_dist(
'normal', **{
'loc': loc,
'scale': scale
})
pass
def test_supp_dist_uni(self):
print(
"========== testing distributions.supported_distributions for Uniform ==========\n"
)
assert_equals(
type(distributions.supported_distributions("Uniform")),
_continuous_distns.uniform_gen,
)
assert_equals(
type(distributions.supported_distributions("U")),
_continuous_distns.uniform_gen,
)
assert_equals(
type(distributions.supported_distributions("Uni")),
_continuous_distns.uniform_gen,
)
def test_supp_dist_norm(self):
print(
"========== testing distributions.supported_distributions for Normal ==========\n"
)
assert_equals(
type(distributions.supported_distributions("Normal")),
_continuous_distns.norm_gen,
)
assert_equals(
type(distributions.supported_distributions("Gauss")),
_continuous_distns.norm_gen,
)
assert_equals(
type(distributions.supported_distributions("Gaussian")),
_continuous_distns.norm_gen,
)
assert_equals(
type(distributions.supported_distributions("Norm")),
_continuous_distns.norm_gen,
)
assert_equals(
type(distributions.supported_distributions("N")),
_continuous_distns.norm_gen,
)
def test_generate_samples(self):
print(
'\n========== testing `sample.sample_set.generate_samples` ==========\n'
)
D = distributions.supported_distributions(
) # dictionary of distributions
for n in [100, 200, 500]: # num_samples
for d in [1, 2, 3]: # dim
self.S.set_num_samples(n)
self.S.set_dim(d)
for dist_key in D.keys():
if dist_key not in ['chi2', 'beta', 'gamma']:
self.S.set_dist(dist_key)
self.S.generate_samples(self.S.num_samples)
# ensure the samples were generated with the correct shape
assert self.S.samples.shape == (n, d)
def test_set_dist(
self): # this essentially functions as a test of assign_dist
print('\n========== testing sample.sample_set.set_dist` ==========\n')
self.S.setup()
D = distributions.supported_distributions(
) # dictionary of distributions
# test `self.S.set_dist()` with no arguments.
# print(type(self.S.dist))
assert type(self.S.dist) is _distn_infrastructure.rv_frozen
for dist_key in D.keys(
): # the dist_keys are strings representing the distributions supported.
# self.S.set_dist(dist_key)
# assert type(self.S.dist) is type(D[dist_key]()) # check that it got instantiated to the expected type.
# now check that argument parameters are passed, stored, and recovered correctly
sz = (4, 5) # test argument passing
if dist_key in ['normal', 'uniform']:
for (l, s) in zip(range(0, 5), range(1, 6)):
self.S.set_dist(dist_key, {'loc': l, 'scale': s})
kwd_dict = self.S.dist.kwds
loc = kwd_dict['loc']
scale = kwd_dict['scale']
assert (loc == l and scale == s)
# assert self.S.dist.args == sz
if dist_key in ['chi2']:
for df in range(1):
self.S.set_dist(dist_key, {'df': df})
kwd_dict = self.S.dist.kwds
deg_freedom = kwd_dict['df']
assert df == deg_freedom
if dist_key in ['gamma']:
for a in range(1):
self.S.set_dist(dist_key, {'a': a})
kwd_dict = self.S.dist.kwds
shape = kwd_dict['a']
assert a == shape
if dist_key in ['beta']:
for (a, b) in zip(range(0, 5), range(1, 6)):
self.S.set_dist(dist_key, {'a': a, 'b': b})
kwd_dict = self.S.dist.kwds
aa = kwd_dict['a']
bb = kwd_dict['b']
assert (a == aa and b == bb)