def test_vis_sample():
msname = os.path.join(os.path.dirname(__file__), 'testfile.ms')
#imagename = os.path.join(os.path.dirname(__file__), 'gaussiantestmodel.fits')
imagename = os.path.join(os.path.dirname(__file__), 'model.fits')
data_vis = import_data_ms(msname)
interp_vis = vis_sample(imagefile = imagename, uvfile = msname, mu_RA = 0, mu_DEC = 0)
print(interp_vis[10:20])
print(uvgaussian(data_vis.uu[10:20], data_vis.vv[10:20]))
def gen_mdl(pars, name):
# calculate a model image
foo = img_parser(inc=pars[0],
PA=pars[1],
dist=140.,
r0=10.,
r_l=pars[2],
z0=pars[3],
zpsi=pars[4],
zphi=np.inf,
Tb0=pars[5],
Tbq=pars[6],
Tbeps=np.inf,
Tbmax=500.,
Tbmax_b=500.,
FOV=2.56,
Npix=512)
# load the template visibilities
dat = np.load('data/template.vis.npz')
u, v, vis, wgt = dat['u'], dat['v'], dat['Vis'], dat['Wgt']
# FT and sample the model image onto the (u,v) spacings
mvis = vis_sample(imagefile=foo,
uu=u,
vv=v,
mu_RA=pars[7],
mu_DEC=pars[8],
mod_interp=False)
print(mvis.shape)
# Save the model and residual visibilities
os.system('rm -rf data/' + name + '.vis.npz')
np.savez('data/' + name + '.vis.npz',
u=u,
v=v,
Vis=np.squeeze(mvis),
Wgt=wgt)
def lnprob(theta):
# calculate prior
ptheta = np.empty_like(theta)
ptheta[0] = lnpU(theta[0], 0., 90.)
ptheta[1] = lnpU(theta[1], 0., 360.)
ptheta[2] = lnpU(theta[2], 0., 5.)
ptheta[3] = lnpU(theta[3], r0, 0.5 * (dist * FOV))
ptheta[4] = lnpU(theta[4], 0., 10.)
ptheta[5] = lnpU(theta[5], 0., 1.5)
ptheta[6] = lnpU(theta[6], 5., Tbmax)
ptheta[7] = lnpU(theta[7], 0., 2.)
ptheta[8] = lnpU(theta[8], 5., 50.)
ptheta[9] = lnpU(theta[9], 0., 1000.)
ptheta[10] = lnpU(theta[10], 3.5, 4.5)
ptheta[11] = lnpU(theta[11], -0.2, 0.2)
ptheta[12] = lnpU(theta[12], -0.2, 0.2)
lnprior = np.sum(ptheta)
# generate a model cube
mcube = cube_parser(inc=theta[0],
PA=theta[1],
dist=dist,
r0=r0,
mstar=theta[2],
r_l=theta[3],
z0=theta[4],
zpsi=theta[5],
Tb0=theta[6],
Tbq=theta[7],
Tbmax=Tbmax,
Tbmax_b=theta[8],
dV0=theta[9],
dVq=0.5 * theta[7],
FOV=FOV,
Npix=Npix,
Vsys=theta[10],
restfreq=nu_l,
vel=v_LSRK_mid)
# sample the FT of the cube onto the observed (u,v) points
mvis = vis_sample(imagefile=mcube,
mu_RA=theta[11],
mu_DEC=theta[12],
gcf_holder=gcf,
corr_cache=corr,
mod_interp=False)
# window the visibilities
SRF_kernel = np.array([0.0, 0.25, 0.5, 0.25, 0.0])
mvis_re = convolve1d(mvis.real, SRF_kernel, axis=1, mode='nearest')
mvis_im = convolve1d(mvis.imag, SRF_kernel, axis=1, mode='nearest')
mvis = mvis_re + 1.0j * mvis_im
# interpolation
fint = interp1d(freq_LSRK_mid, mvis, axis=1, fill_value='extrapolate')
mvis = fint(data.freqs)
# excise the padded boundary channels to avoid edge effects
mvis = mvis[:, chpad:-chpad].T
mwgt = data.wgts[chpad:-chpad, :]
# clip for binning
mvis = mvis[:mvis.shape[0] - (mvis.shape[0] % chbin), :]
mwgt = mwgt[:mvis.shape[0] - (mvis.shape[0] % chbin), :]
# bin (weighted, decimated average)
mvis_bin = np.average(mvis.reshape((-1, chbin, mvis.shape[1])),
weights=mwgt.reshape((-1, chbin, mwgt.shape[1])),
axis=1)
# compute the log-likelihood
resid = np.absolute(data_bin.VV - mvis_bin)
lnL = -0.5 * np.tensordot(resid, np.dot(Mbin_inv, data_bin.wgts * resid))
# return the posterior
return lnL + L0 + lnprior, lnprior
zpsi=theta[5],
Tb0=theta[6],
Tbq=theta[7],
Tbmax=Tbmax,
Tbmax_b=theta[8],
dV0=theta[9],
dVq=0.5 * theta[7],
FOV=FOV,
Npix=Npix,
Vsys=theta[10],
restfreq=nu_l,
vel=v_LSRK_mid)
tvis, gcf, corr = vis_sample(imagefile=foo,
uu=data.uu,
vv=data.vv,
return_gcf=True,
return_corr_cache=True,
mod_interp=False)
### PRIOR FUNCTIONAL FORMS
### ----------------------
# uniform
def lnpU(theta, lo, hi):
if ((theta >= lo) and (theta <= hi)):
return 0
else:
return -np.inf
# normal
Tb0=theta[6],
Tbq=theta[7],
Tbmax=Tbmax,
Tbmax_b=theta[8],
dV0=theta[9],
dVq=theta[10],
FOV=FOV,
Npix=Npix,
Vsys=theta[11],
restfreq=restfreq,
vel=1e3 * v_LSRK[i, :])
# sample its Fourier Transform on the template (u,v) spacings
mvis = vis_sample(imagefile=foo,
uu=tvis.uu,
vv=tvis.vv,
mu_RA=theta[12],
mu_DEC=theta[13],
mod_interp=False)
# populate the appropriate parts of cloned array with these visibilities
ix_lo, ix_hi = i * nperstamp, (i + 1) * nperstamp
clean_arr[ix_lo:ix_hi, :, 0, 0] = mvis.real[ix_lo:ix_hi, :]
clean_arr[ix_lo:ix_hi, :, 1, 0] = mvis.real[ix_lo:ix_hi, :]
clean_arr[ix_lo:ix_hi, :, 0, 1] = mvis.imag[ix_lo:ix_hi, :]
clean_arr[ix_lo:ix_hi, :, 1, 1] = mvis.imag[ix_lo:ix_hi, :]
# the same, but with noise added
noisy_arr[ix_lo:ix_hi,:,0,0] = mvis.real[ix_lo:ix_hi,:] + \
noiseXX.real[ix_lo:ix_hi,:]
noisy_arr[ix_lo:ix_hi,:,1,0] = mvis.real[ix_lo:ix_hi,:] + \
noiseYY.real[ix_lo:ix_hi,:]