def reggrad(imvec):
return iu.regularizergrad(imvec,
nprior,
embed_mask,
flux,
im.xdim,
im.ydim,
im.psize,
rtype,
beam_size=20. * eh.RADPERUAS,
alpha_A=alpha_A,
norm_reg=True)
mask=mask)
compare_floor = 1.0
print(
"Median Fractional Difference of DTFT/NFFT for chi^2 gradient of " +
dtype,
np.median(
np.abs((chisq_dft_grad - chisq_nfft_grad) /
(np.abs(chisq_dft_grad) + compare_floor))))
# Testing the gradients of image regularization functions
prior = test_imvec * 0.0 + 1.0
prior = prior * np.sum(test_imvec) / np.sum(prior)
mask = [
True,
] * len(test_imvec)
for reg in ['simple', 'gs', 'l1', 'tv', 'tv2']:
dx = 1.e-12
y0 = iu.regularizer(test_imvec, prior, mask, 1.0, im.xdim, im.ydim,
im.psize, reg)
grad_exact = iu.regularizergrad(test_imvec, prior, mask, 1.0, im.xdim,
im.ydim, im.psize, reg)
grad = np.zeros(len(test_imvec))
for j in range(len(test_imvec)):
test_imvec2 = test_imvec.copy()
test_imvec2[j] += dx
y1 = iu.regularizer(test_imvec2, prior, mask, 1.0, im.xdim, im.ydim,
im.psize, reg)
grad[j] = (y1 - y0) / dx
print("Median Fractional Gradient Difference for " + reg + ":",
np.median(np.abs((grad - grad_exact) / grad_exact)))