def test_tv_rw_sdmm():
from numpy.linalg import norm
from numpy.testing import assert_allclose
from numpy import sqrt
from purify.tests.random import reset
from purify import SDMM
from purify.tests.image_fixtures import image_and_visibilities, \
expected_images, dirty_measurements
image, visibility = image_and_visibilities()
expected = expected_images('30dor', 'tvrw')
reset()
wavelets = ['DB%i' % i for i in range(1, 9)] + ['Dirac']
rwsdmm = SDMM(image_size=image.shape,
nblevels=4,
wavelets=wavelets,
reweighted=True,
tv_norm=True)
visibility['y'] = dirty_measurements(image, visibility)
sigma = norm(visibility['y0']) * 10.0**(-1.5) / sqrt(len(visibility))
radius = sqrt(1e0 + 2.0 / sqrt(len(visibility))) * sigma \
* sqrt(image.size)
rwsdmm.rw.sigma = sigma * sqrt(len(visibility)) \
/ sqrt(len(rwsdmm) * image.size)
rwsdmm.rw.max_iter = 2
rwsdmm.relative_variation = 1e-3
rwsdmm.rw.relative_variation = 1e-3
rwsdmm.rw.warm_start = False
actual = rwsdmm(visibility, radius=radius, max_iter=3)
assert_allclose(actual.real, expected)
def test_tv_sdmm():
from numpy.linalg import norm
from numpy.testing import assert_allclose
from numpy import sqrt
from purify.tests.random import reset
from purify import SDMM
from purify.tests.image_fixtures import image_and_visibilities, \
expected_images, dirty_measurements
image, visibility = image_and_visibilities()
expected = expected_images('30dor', 'tv')
reset()
sdmm = SDMM(image_size=image.shape)
visibility['y'] = dirty_measurements(image, visibility)
sigma = norm(visibility['y0']) * 10.0**(-1.5) / sqrt(len(visibility))
radius = sqrt(1e0 + 2.0 / sqrt(len(visibility))) * sigma \
* sqrt(image.size)
actual = sdmm(visibility,
radius=radius,
max_iter=6,
tv_norm=True,
relative_variation=1e-3)
assert_allclose(actual.real, expected)
def test_harden_random():
""" Checks that random numbers are fixed by seed
These numbers are used to generate fake dirty images from cleaned
images in both C and python code. They should remain the same.
"""
from numpy.testing import assert_allclose
from purify.tests.random import reset
image, visibility = image_and_visibilities()
reset()
actual = make_some_noise(visibility, 5)
expected = [
-0.0004048868413905625 + 0.0004998856543791353j,
0.0010252113233472518 + 0.0006731309681747155j,
-0.0006402841370415399 - 0.000710428273313625j,
-0.0015624288645055503 - 0.0016640991757716221j,
0.0015712341522236288 + 0.0008254357025121923j
]
assert_allclose(actual, expected)
def test_l1_sdmm():
from numpy.linalg import norm
from numpy.testing import assert_allclose
from numpy import sqrt
from purify.tests.random import reset
from purify import SDMM
from purify.tests.image_fixtures import image_and_visibilities, \
expected_images, dirty_measurements
image, visibility = image_and_visibilities()
expected = expected_images('30dor')
reset()
wavelets = ['DB%i' % i for i in range(1, 9)] + ['Dirac']
sdmm = SDMM(image_size=image.shape, nblevels=4, wavelets=wavelets)
visibility['y'] = dirty_measurements(image, visibility)
sigma = norm(visibility['y0']) * 10.0**(-1.5) / sqrt(len(visibility))
radius = sqrt(1e0 + 2.0 / sqrt(len(visibility))) * sigma \
* sqrt(image.size)
actual = sdmm(visibility, radius=radius, max_iter=5)
assert_allclose(actual.real, expected)
def test_harden_dirty_image():
""" Make sure dirty image stays what it is. """
from numpy.testing import assert_allclose
from numpy import sqrt
from purify.tests.random import reset
image, visibility = image_and_visibilities()
reset()
actual = dirty_measurements(image, visibility)[:10]
expected = [
(-0.0025779629409+0.000750385437798j),
(0.00114034979371-0.00100676562853j),
(-0.00136686865551-0.00159946686582j),
(-0.00198922875103+0.00733872190704j),
(-0.0060053359212-0.000512419375827j),
(0.00266958514619-0.00375981242741j),
(0.00690335501097-0.00766305939611j),
(-0.000400339081838+0.00084899489691j),
(0.0034246920022-0.0124356716948j),
(0.000650022911397+0.0005080970237j),
]
scale = sqrt(image.size) / sqrt(len(visibility))
assert_allclose(actual, expected / scale)