datai[i] = im[i][mask == True]
datai_mean = stats.compute_mean(datai, n_i, col)
dataii = copy.copy(datai)
dataii = stats.remove_mean(dataii, datai_mean, n_i, col)
displ = copy.copy(datai)
for i in range(n_i):
mask_gaps[i][im[i][mask == True] == 0] = True
mask_g_temp[i] = copy.copy(mask_gaps[i])
mask_gaps[i] = gg.gen_cv_mask(mask_gaps[i], n_pts, n_cv[0])
mask_cv = np.logical_xor(mask_gaps, mask_g_temp)
# Apply mask on displacement field
displ = gg.mask_field(displ, mask_gaps, np.nan)
displ_mean = stats.compute_mean(displ, n_i, col)
displ = stats.remove_mean(displ, displ_mean, n_i, col)
displtp = copy.copy(displ)
# Initialization
NOISE_TYPE = 'rand'
mu, sigma = 0, 1
blanc = np.random.normal(mu, sigma, len(displtp[mask_gaps == True]))
if NOISE_TYPE == 'corr':
expo = 1.4 # exponent in correlation function (as: 1/(r)**expo)
geo_noise = ng.geo(r, expo)
noise = ng.gen_noise_series(geo_noise, blanc, nt)
elif NOISE_TYPE == 'rand':
noise = blanc
for m in range(len(ngaps)):
tng = ngaps[m] * nt
mask_temp = copy.copy(mask0)
for i in range(nt):
mask_temp[:, i] = gg.gen_cv_mask(mask_temp[:, i], nobs,
ngaps_cv[m])
# Generate mask for cross validation
mask_cv = np.logical_xor(mask_temp, mask0)
# Create mask where data exists for later use
mask_data = np.invert(mask_temp)
n_data = len(mask_data[mask_data == True])
# Apply mask on displacement field
fdispl = gg.mask_field(fdispl, mask_temp, np.nan)
# Create a blank image
#fdispl[:,10][1000:30000] = np.nan
displ_mean = compute_mean(fdispl, col)
fdispl = remove_mean(fdispl, displ_mean, nt, col)
noise_init = np.random.normal(mu, sigma, (nt, nx, ny))
noise_init *= 8
fval = [0.0, np.reshape(noise_init, (nt, nobs)).T]
for init in inits: # loop for type of gap initialization
# Gap initialization
fdispltp = copy.deepcopy(fdispl)
if init == 'zero':
fdispltp[mask_temp == True] = fval[0]