shuffle(pairs)
else:
raise NotImplementedError
# Sampling points
points = pairs.points
# TODO Add design points; e.g. corners of the plot
#from plotting import lllat, lllon, urlat, urlon
#grid = np.rec.fromarrays(np.mgrid[lllat:urlat:5j, lllon:urlon:5j], dtype=dt_latlon)
#points = np.concatenate( (pairs.points, grid.flatten()))
# A priori assumptions
mu_C = np.full_like(points, mu,
dtype=np.float) # The velocity models a priori mean
# A priori covariance
cov_CC = gauss_kernel(points[:, np.newaxis], points[np.newaxis, :], tau, ell)
# Open HDF5 file handle
output_file = config.get('Output', 'filename')
fh = h5py.File(output_file, 'w')
# Store stations
fh.create_dataset('stations', data=pairs.stations)
# Store discretization
fh.create_dataset('points', data=points)
# Store prior covariance matrix
fh.create_dataset('cov_CC_pri', data=cov_CC)
# Create datasets for mean, standard deviation and misfit
dset_mu = fh.create_dataset('mu', (len(pairs) + 1, ) + mu_C.shape)
dset_sd = fh.create_dataset('sd', (len(pairs) + 1, ) + mu_C.shape)
increment = 20
shp_misfit = (max(1, len(pairs) / increment) + 1, )
def __call__(self, crd):
res = np.full_like(crd, self._v0, dtype=np.float)
for v, x, r in zip(self._vs, self._xs, self._rs):
res += v * gauss_kernel(crd, x, ell=r, tau=1)
return res
# Open HDF5 file handle
output_file = config.get('Output', 'filename')
fh = h5py.File(output_file, 'w')
# Store stations
fh.create_dataset('stations', data=pairs.stations)
# Get discretization
points = pairs.points
# Store discretization
fh.create_dataset('points', data=points)
# A priori mean and assumed covariance
mu_C = np.full_like(points, mu, dtype=np.float)
cov_CC = gauss_kernel(points.reshape(1, -1),
points.reshape(-1, 1),
ell=ell,
tau=tau)
# Prior predictive mean and covariance
mu_D_pri = pairs.mu_T(mu_C)
cov_DD_pri = pairs.cov_TT(mu_C, cov_CC)
# Store prior assumptions
dset_mu = fh.create_dataset('mu', (2, ) + mu_C.shape)
dset_mu[0, :] = mu_C
dset_sd = fh.create_dataset('sd', (2, ) + mu_C.shape)
dset_sd[0, :] = np.sqrt(cov_CC.diagonal())
fh.create_dataset('cov_DD_pri', data=cov_DD_pri)
# Store prior misfit
dt_misfit = np.dtype([('evd', np.int), ('val', float)])
dset_misfit = fh.create_dataset('misfit', (2, ), dtype=dt_misfit)
dset_misfit[0] = 0, pairs.misfit(mu_C, cov_CC)
zorder=10)
# Find stations which are the furtherest apart
pair = max(pairs, key=lambda pair: pair.central_angle)
# Middle point of the great circle path
p = pair.great_circle_path[12]
# Make a lat, lon grid round the middle point
N = 100j
lllat = p['lat'] - 1
urlat = p['lat'] + 1
lllon = p['lon'] - 1.5
urlon = p['lon'] + 1.5
grid = np.rec.fromarrays(np.mgrid[lllat:urlat:N, lllon:urlon:N],
dtype=dt_latlon)
# Calculate kernel at the middle point
K = gauss_kernel(p, grid, tau=tau, ell=ell)
K = np.ma.masked_less(K, K.max() / 50)
# Plot correlation kernel; pcolor needs points in between
lat, lon = np.mgrid[lllat:urlat:N + 1j, lllon:urlon:N + 1j]
pcol = m.pcolormesh(lon, lat, K, latlon=True, cmap=cmap_sd, rasterized=True, \
vmin=0, vmax=K.max(), zorder=1)
# Make colorbar
cbar = plt.colorbar(pcol, cax=ax_cbr, orientation='horizontal')
cbar.set_label(r'$m^2 s^{-2}$')
cbar.solids.set_edgecolor("face")
plt.savefig('../fig_kernel_pri.pgf')
# Find stations which are the furtherest apart
pair = max(pairs, key=lambda pair: pair.central_angle)
# Parametrization of the great circle path
path = pair.great_circle_path
# Make a lat, lon grid with the extent of the path
N = 200j
lllat = min(pair.st1['lat'], pair.st2['lat']) - 0.5
urlat = max(pair.st1['lat'], pair.st2['lat']) + 0.5
lllon = min(pair.st1['lon'], pair.st2['lon']) - 1
urlon = max(pair.st1['lon'], pair.st2['lon']) + 1
grid = np.rec.fromarrays(np.mgrid[lllat:urlat:N, lllon:urlon:N],
dtype=dt_latlon)
# Calculate Correlations amongst great circle segment and grid
K = gauss_kernel(path.reshape((-1, 1, 1)), grid, tau=tau, ell=ell)
# Highlight parametrization of the great circle segment
m.plot(path['lon'], path['lat'], latlon=True, lw=0.5, color='g')
m.scatter(path['lon'][1:-1],
path['lat'][1:-1],
latlon=True,
s=1,
marker='.',
color='g')
xr, yr = m(path[-1]['lon'], path[-1]['lat'])
xs, ys = m(path[0]['lon'], path[0]['lat'])
ax_map.text(xr,
yr,
'r',
fontsize=12,