def test_redcal(self):
#check that logcal give 0 chi2 for all 20 testinfos
for index in xrange(20):
arrayinfopath = os.path.dirname(
os.path.realpath(__file__)) + '/testinfo/test' + str(
index + 1) + '_array_info.txt'
c = Oc.RedundantCalibrator(56)
c.compute_redundantinfo(arrayinfopath, tol=.1)
info = c.Info
npz = np.load(testdata % index)
bls = [tuple(bl) for bl in npz['bls']]
dd = dict(zip(bls, npz['vis']))
m, g, v = Oc.redcal(dd,
info,
removedegen=True,
maxiter=50,
stepsize=.2,
computeUBLFit=True,
conv=1e-5,
uselogcal=True)
calparpath = os.path.dirname(os.path.realpath(
__file__)) + '/testinfo/test' + str(index + 1) + '_calpar.txt'
with open(calparpath) as f:
rawinfo = [[float(x) for x in line.split()] for line in f]
temp = np.array(rawinfo[:-1])
correctcalpar = (np.array(temp[:, 0]) +
1.0j * np.array(temp[:, 1]))
i = g.keys()[0]
scalar = correctcalpar[i].real / g[i].real
for i in xrange(56):
if not g.has_key(i): continue
self.assertAlmostEqual(
np.abs(correctcalpar[i] - g[i] * scalar), 0, 4)
def test_large_info_IO(self):
calibrator = Oc.RedundantCalibrator(150)
calibrator.compute_redundantinfo()
calibrator.write_redundantinfo(infotestpath, verbose=VERBOSE)
info2 = Oi.RedundantInfo(filename=infotestpath)
self.assertEqual(calibrator.Info.nAntenna, info2.nAntenna)
self.assertEqual(calibrator.Info.nBaseline, info2.nBaseline)
self.assertEqual(calibrator.Info.get_reds(), info2.get_reds())
os.remove(infotestpath)
def test_logcal(self):
#check that logcal give 0 chi2 for all 20 testinfos
diff = np.zeros(20)
for index in range(20):
arrayinfopath = os.path.dirname(
os.path.realpath(__file__)) + '/testinfo/test' + str(
index + 1) + '_array_info.txt'
calibrator = Oc.RedundantCalibrator(56)
calibrator.compute_redundantinfo(arrayinfopath, tol=.1)
if False: # XXX this was to migrate files so they include bl order w/ data
_info = calibrator.Info # XXX needs to have been initialized the old way (w/o reds)
datapath = os.path.dirname(
os.path.realpath(__file__)) + '/testinfo/test' + str(
index + 1) + '_data.txt'
with open(datapath) as f:
rawinfo = [[float(x) for x in line.split()] for line in f]
data = np.array([i[0] + 1.0j * i[1] for i in rawinfo[:-1]],
dtype='complex64') #last element is empty
data = data.reshape((1, 1, len(data)))
dd = _info.make_dd(data)
np.savez('calib_test_data_%02d.npz' % index,
bls=np.array(dd.keys()),
vis=np.array(dd.values()))
info = calibrator.Info
npz = np.load(testdata % index)
bls = [tuple(bl) for bl in npz['bls']]
dd = dict(zip(bls, npz['vis']))
data = info.order_data(dd)
####do calibration################
calibrator.removeDegeneracy = True
calibrator.removeAdditive = False
calibrator.keepData = True
calibrator.keepCalpar = True
calibrator.convergePercent = 1e-5
calibrator.maxIteration = 50
calibrator.stepSize = .2
calibrator.computeUBLFit = True
calibrator.logcal(data, np.zeros_like(data), verbose=VERBOSE)
log = np.copy(calibrator.rawCalpar)
ampcal = log[0, 0, 3:info['nAntenna'] + 3]
phasecal = log[0, 0, info['nAntenna'] + 3:info['nAntenna'] * 2 + 3]
calpar = 10**(ampcal) * np.exp(1.0j * phasecal)
ublfit = log[0, 0, 3 + 2 * info['nAntenna']::2] + 1.0j * log[
0, 0, 3 + 2 * info['nAntenna'] + 1::2]
####import real calibration parameter
calparpath = os.path.dirname(os.path.realpath(
__file__)) + '/testinfo/test' + str(index + 1) + '_calpar.txt'
with open(calparpath) as f:
rawinfo = [[float(x) for x in line.split()] for line in f]
temp = np.array(rawinfo[:-1])
correctcalpar = (np.array(temp[:, 0]) +
1.0j * np.array(temp[:, 1]))[info['subsetant']]
###compare calpar with correct calpar
overallfactor = np.real(np.mean(ublfit))**0.5
diffnorm = la.norm(calpar * overallfactor - correctcalpar)
self.assertAlmostEqual(diffnorm, 0, 4)
def test_lincal(self):
fileindex = 3 #use the 3rd file to do the test, can also change this to any number from 1 to 20
length = 100
loglist = np.zeros(length)
linlist = np.zeros(length)
####import arrayinfo################
arrayinfopath = os.path.dirname(os.path.realpath(
__file__)) + '/testinfo/test' + str(fileindex) + '_array_info.txt'
nant = 56
calibrator = Oc.RedundantCalibrator(nant)
calibrator.compute_redundantinfo(arrayinfopath)
info = calibrator.Info
npz = np.load(testdata % (fileindex - 1))
bls = [tuple(bl) for bl in npz['bls']]
dd = dict(zip(bls, npz['vis']))
data = info.order_data(dd)
####Config parameters###################################
needrawcal = True #if true, (generally true for raw data) you need to take care of having raw calibration parameters in float32 binary format freq x nant
std = 0.1
####do calibration################
calibrator.removeDegeneracy = True
calibrator.removeAdditive = False
calibrator.keepData = True
calibrator.keepCalpar = True
calibrator.convergePercent = 1e-5
calibrator.maxIteration = 50
calibrator.stepSize = .2
calibrator.computeUBLFit = True
for i in range(length):
noise = (np.random.normal(scale=std, size=data.shape) +
1.0j * np.random.normal(scale=std, size=data.shape)
).astype('complex64')
ndata = data + noise
calibrator.logcal(ndata, np.zeros_like(ndata), verbose=VERBOSE)
calibrator.lincal(ndata, np.zeros_like(ndata), verbose=VERBOSE)
linchi2 = (calibrator.rawCalpar[0, 0, 2] /
(info['At'].shape[1] - info['At'].shape[0]) /
(2 * std**2))**0.5
logchi2 = (calibrator.rawCalpar[0, 0, 1] /
(info['At'].shape[1] - info['At'].shape[0]) /
(2 * std**2))**0.5
linlist[i] = linchi2
loglist[i] = logchi2
self.assertTrue(abs(np.mean(linlist) - 1.0) <
0.01) #check that chi2 of lincal is close enough to 1
self.assertTrue(
np.mean(linlist) < np.mean(loglist)
) #chick that chi2 of lincal is smaller than chi2 of logcal
def test_lincal(self):
fileindex = 3 #use the 3rd file to do the test, can also change this to any number from 1 to 20
length = 100
loglist = np.zeros(length)
linlist = np.zeros(length)
####import arrayinfo################
arrayinfopath = os.path.dirname(os.path.realpath(
__file__)) + '/testinfo/test' + str(fileindex) + '_array_info.txt'
nant = 56
calibrator = Oc.RedundantCalibrator(nant)
calibrator.compute_redundantinfo(arrayinfopath)
info = calibrator.Info
npz = np.load(testdata % (fileindex - 1))
bls = [tuple(bl) for bl in npz['bls']]
dd = dict(zip(bls, npz['vis']))
data = np.array([
dd[bl] if bl in dd else dd[bl[::-1]].conj()
for bl in info.bl_order()
]).transpose((1, 2, 0))
#data = info.order_data(dd)
####Config parameters###################################
needrawcal = True #if true, (generally true for raw data) you need to take care of having raw calibration parameters in float32 binary format freq x nant
std = 0.1
####do calibration################
calibrator.removeDegeneracy = True
calibrator.removeAdditive = False
calibrator.keepData = True
calibrator.keepCalpar = True
calibrator.convergePercent = 1e-5
calibrator.maxIteration = 50
calibrator.stepSize = .2
calibrator.computeUBLFit = True
for i in range(length):
noise = (np.random.normal(scale=std, size=data.shape) +
1.0j * np.random.normal(scale=std, size=data.shape)
).astype('complex64')
ndata = data + noise
calibrator.logcal(ndata, np.zeros_like(ndata), verbose=VERBOSE)
calibrator.lincal(ndata, np.zeros_like(ndata), verbose=VERBOSE)
linchi2 = (calibrator.rawCalpar[0, 0, 2] /
(calibrator.Info['At'].shape[1] -
calibrator.Info['At'].shape[0]) / (2 * std**2))**0.5
logchi2 = (calibrator.rawCalpar[0, 0, 1] /
(calibrator.Info['At'].shape[1] -
calibrator.Info['At'].shape[0]) / (2 * std**2))**0.5
linlist[i] = linchi2
loglist[i] = logchi2
# The sum of the chi^2's for all antennas should be twice the
# calibration chi^2. Omnical uses single precision floats in C, so
# the ratio of that sum to chi^2 must be equal to 2 to 5 decimal
# places. That should hold in all cases.
chi2all = calibrator.rawCalpar[0, 0, 2]
chi2ant = calibrator.rawCalpar[
0, 0, 3 + 2 * (info.nAntenna + len(info.ublcount)):]
np.testing.assert_almost_equal(np.sum(chi2ant) / chi2all - 2, 0, 5)
self.assertTrue(abs(np.mean(linlist) - 1.0) <
0.01) #check that chi2 of lincal is close enough to 1
self.assertTrue(
np.mean(linlist) < np.mean(loglist)
) #chick that chi2 of lincal is smaller than chi2 of logcal
