def RMclean1(fq,
spec,
gain=0.1,
tol=1e-3,
stop_if_div=False,
maxiter=10000,
verbose=True,
pos_def=False,
window='hamming'):
RMs, W = RMTmat(fq, window)
rm_spec = RMT(spec, W)
ker_spec = RMT(np.ones_like(spec), W)
mod, info = deconv.clean(rm_spec,
ker_spec,
gain=gain,
maxiter=maxiter,
tol=tol,
stop_if_div=stop_if_div,
verbose=verbose)
W = iRMTmat(fq, window)
mod_spec = iRMT(mod, W)
res_spec = iRMT(info['res'], W)
return mod_spec, res_spec
def test_tolerances(self):
A0 = deconv.clean(img0, ker, stop_if_div=False, tol=0.01)[0]
A1 = deconv.clean(img1, ker, stop_if_div=False, tol=1e-6)[0]
A2 = deconv.clean(img2, ker, stop_if_div=False, tol=1e-9)[0]
B0 = clean(img0, ker, stop_if_div=False, tol=0.01)[0]
B1 = clean(img1, ker, stop_if_div=False, tol=1e-6)[0]
B2 = clean([img2] * 3, [ker] * 3, stop_if_div=False, tol=1e-9)[0][1]
for i in xrange(1024):
self.assertEqual(A0[i], B0[i])
for i in xrange(1024):
self.assertEqual(A1[i], B1[i])
for i in xrange(1024):
self.assertEqual(A2[i], B2[i])
def test_many(self):
A0 = deconv.clean(img0, ker0, stop_if_div=False, tol=0)[0]
A1 = deconv.clean(img1, ker1, stop_if_div=False, tol=0)[0]
A2 = deconv.clean(img2, ker0, stop_if_div=False, tol=0)[0]
A3 = deconv.clean(img3, ker1, stop_if_div=False, tol=0)[0]
A4 = deconv.clean(img4, ker0, stop_if_div=False, tol=0)[0]
A5 = deconv.clean(img5, ker1, stop_if_div=False, tol=0)[0]
A6 = deconv.clean(img6, ker0, stop_if_div=False, tol=0)[0]
A7 = deconv.clean(img7, ker1, stop_if_div=False, tol=0)[0]
A8 = deconv.clean(img8, ker0, stop_if_div=False, tol=0)[0]
B = clean(imgs, kers, stop_if_div=False, tol=0)[0]
for i in xrange(1024):
self.assertAlmostEqual(A5[i], B[5][i])
def test_RandomInput(self):
dim = 25
img = np.array(np.random.rand(dim), dtype=np.float32)
ker = np.array(np.random.rand(dim), dtype=np.float32)
A = deconv.clean(img, ker)[0]
B = clean(img, ker)[0]
for i in xrange(dim):
self.assertEqual(A[i], B[i])
def test_mdls(self):
A0 = deconv.clean(img0,
ker,
np.array(area0, dtype=np.complex64),
stop_if_div=False,
tol=0.1,
verbose=False)[0]
A1 = deconv.clean(img1,
ker,
stop_if_div=False,
tol=1e-6,
mdl=np.array(area1, dtype=np.complex64),
verbose=False)[0]
A2 = deconv.clean(img2,
ker,
stop_if_div=False,
tol=1e-9,
mdl=np.array(area2, dtype=np.complex64),
verbose=False)[0]
B0 = clean(img0,
ker,
stop_if_div=False,
tol=0.1,
mdl=np.array(area0, dtype=np.complex64))[0]
B1 = clean(img1,
ker,
stop_if_div=False,
tol=1e-6,
mdl=np.array(area1, dtype=np.complex64))[0]
B2 = clean([img2] * 3, [ker] * 3,
stop_if_div=False,
tol=1e-9,
mdl=[np.array(area2, dtype=np.complex64)] * 3)[0][1]
for i in xrange(1024):
self.assertEqual(A0[i], B0[i])
for i in xrange(1024):
self.assertEqual(A1[i], B1[i])
for i in xrange(1024):
self.assertEqual(A2[i], B2[i])
def test_Default(self):
img = np.array([0, 0, 0, 4, 6, 4, 0, 0, -2, -3, -2, 0],
dtype=np.float32)
ker = np.array([3, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2], dtype=np.float32)
#print deconv.clean(img, ker, verbose=True)[0]
#print clean(img, ker)[0]
for i in xrange(12):
self.assertEqual(
deconv.clean(img, ker)[0][i],
clean(img, ker)[0][i])
def test_stop_if_div(self):
A0 = deconv.clean(img0, ker, stop_if_div=True, tol=0.1)[0]
A1 = deconv.clean(img1, ker, stop_if_div=True, tol=1e-6)[0]
A2 = deconv.clean(img2, ker, stop_if_div=True, tol=1e-9)[0]
A3 = deconv.clean(img3, ker128, stop_if_div=True, tol=1e-6)[0]
B0 = clean(img0, ker, stop_if_div=True, tol=0.1)[0]
B1 = clean(img1, ker, stop_if_div=True, tol=1e-6)[0]
B2 = clean([img2] * 3, [ker] * 3, stop_if_div=True, tol=1e-9)[0][1]
B3 = clean(img3, ker128, stop_if_div=True, tol=1e-6)[0]
for i in xrange(1024):
self.assertEqual(A0[i], B0[i])
for i in xrange(1024):
self.assertEqual(A1[i], B1[i])
for i in xrange(1024):
self.assertEqual(A2[i], B2[i])
for i in xrange(1024):
self.assertEqual(A3[i], B3[i])
def test_spike(self):
ker = np.zeros(1024, dtype=np.float32)
ker[0] = 1
img = ker.copy()
A, info = deconv.clean(img,
ker,
stop_if_div=True,
tol=0,
maxiter=int(1e4))
B = clean(img, ker, stop_if_div=True, tol=0, maxiter=int(1e4))[0]
for i in xrange(1024):
self.assertEqual(A[i], B[i])
def test_foregrounds(self):
fqs = np.linspace(.1, .2, 1024, endpoint=False)
lsts = np.linspace(0, 2 * np.pi, 10000, endpoint=False)
bl_len_ns = 30.
vis_fg_pntsrc = foregrounds.pntsrc_foreground(lsts,
fqs,
bl_len_ns,
nsrcs=200)
img0 = np.array(vis_fg_pntsrc[100], dtype=np.float32)
img1 = np.array(vis_fg_pntsrc[500], dtype=np.float32)
img2 = np.array(vis_fg_pntsrc[700], dtype=np.float32)
ker = np.ones(1024)
A = set()
while len(A) < 160:
A.add(random.randint(0, 1024))
for i in xrange(len(ker)):
if i in A:
ker[i] = 0
ker = np.array(np.fft.fft(ker), dtype=np.float32)
A0, info = deconv.clean(img0,
ker,
stop_if_div=True,
tol=0,
verbose=True)
# A1 = deconv.clean(img1, ker, stop_if_div=True, tol=0)[0]
# A2 = deconv.clean(img2, ker, stop_if_div=True, tol=0)[0]
B0 = clean(img0, ker, stop_if_div=True, tol=0)[0]
# B1 = clean(img1, ker, stop_if_div=True, tol=0)[0]
# B2 = clean([img2]*3, [ker]*3, stop_if_div=False, tol=0)[0][1]
for i in xrange(1024):
self.assertEqual(A0[i], B0[i])
ker = np.fft.fft(ker)
ker = np.array(ker, dtype=data_type)
numImgs = 10
# numImgs = 4096
print numImgs, data_type, "Images"
print "AIPY CLEAN"
s = cuda.Event()
e = cuda.Event()
s.record()
for i in xrange(numImgs):
deconv.clean(img1, ker, stop_if_div=False, tol=0)
e.record()
e.synchronize()
print s.time_till(e) / 1000, "s"
s = cuda.Event()
e = cuda.Event()
s.record()
deconv.clean(img2, ker, stop_if_div=False, tol=0)
e.record()
e.synchronize()
print s.time_till(e) / 1000, "s"
# s = cuda.Event()
# e = cuda.Event()
# s.record()