def _test_gauss(slice_sample_fn, prng):
# sample from N(0,1)
pdf = log_normal(0., 1.)
def sampler(x0, niters):
x = x0
for _ in xrange(niters):
x = slice_sample_fn(pdf, x, 2.0, prng)
yield x
bins = np.linspace(-3, 3, 1000)
smoothing = 1e-5
actual_samples = np.random.normal(size=10000)
actual_hist = hist(actual_samples, bins) + smoothing
actual_hist /= actual_hist.sum()
slice_samples = np.array(list(sampler(1.0, 10000)))
slice_hist = hist(slice_samples, bins) + smoothing
slice_hist /= slice_hist.sum()
# KL-divergence
kldiff = KL_approx(actual_hist, slice_hist, bins[1] - bins[0])
print 'KL:', kldiff
# statistical distance
maxdiff = np.abs(actual_hist - slice_hist).max()
print 'maxdiff:', maxdiff
assert kldiff <= 0.005
assert maxdiff <= 0.005
def test_nich_hp_mu():
N = 1000
Y = np.array([(x, ) for x in np.random.uniform(low=-10, high=10, size=N)],
dtype=[('', np.float32)])
view = numpy_dataview(Y)
grid_min, grid_max, grid_n = -5., 5., 100
_test_scalar_hp_inference(view, log_normal(0., 1.), 0.1, grid_min,
grid_max, grid_n, nich, 'mu')
def test_nich_hp_mu():
N = 1000
Y = np.array([(x,) for x in np.random.uniform(low=-10, high=10, size=N)],
dtype=[('', np.float32)])
view = numpy_dataview(Y)
grid_min, grid_max, grid_n = -5., 5., 100
_test_scalar_hp_inference(view,
log_normal(0., 1.),
0.1,
grid_min,
grid_max,
grid_n,
nich,
'mu')
def test_log_normal():
from microscopes.common.scalar_functions import log_normal
from scipy.stats import norm
import math
mu, sigma2 = 1.5, 3.2
x = 6.3
ours = log_normal(mu, sigma2)
assert_almost_equals(
ours(x), norm.logpdf(x, loc=mu, scale=math.sqrt(sigma2)))
ours1 = pickle.loads(pickle.dumps(ours))
assert_almost_equals(ours._mu, ours1._mu)
assert_almost_equals(ours._sigma2, ours1._sigma2)
assert_almost_equals(
ours1(x), norm.logpdf(x, loc=mu, scale=math.sqrt(sigma2)))
ours1 = copy.copy(ours)
assert_almost_equals(ours._mu, ours1._mu)
assert_almost_equals(ours._sigma2, ours1._sigma2)
assert_almost_equals(
ours1(x), norm.logpdf(x, loc=mu, scale=math.sqrt(sigma2)))
def test_log_normal():
from microscopes.common.scalar_functions import log_normal
from scipy.stats import norm
import math
mu, sigma2 = 1.5, 3.2
x = 6.3
ours = log_normal(mu, sigma2)
assert_almost_equals(ours(x),
norm.logpdf(x, loc=mu, scale=math.sqrt(sigma2)))
ours1 = pickle.loads(pickle.dumps(ours))
assert_almost_equals(ours._mu, ours1._mu)
assert_almost_equals(ours._sigma2, ours1._sigma2)
assert_almost_equals(ours1(x),
norm.logpdf(x, loc=mu, scale=math.sqrt(sigma2)))
ours1 = copy.copy(ours)
assert_almost_equals(ours._mu, ours1._mu)
assert_almost_equals(ours._sigma2, ours1._sigma2)
assert_almost_equals(ours1(x),
norm.logpdf(x, loc=mu, scale=math.sqrt(sigma2)))