def test_spherical_surface_elements():
xg, yg = np.meshgrid([0., 1., 2., 3.], [4., 5., 6., 3.])
grid = np.array([np.ravel(xg), np.ravel(yg)])
surfel = get_spherical_surface_elements(grid[0], grid[1])
assert (len(surfel) == len(grid[0]))
assert (type(surfel) == np.ndarray)
assert (pytest.approx(surfel[0]) == 12361699247.23782)
def __init__(self,model,w='r'):
# Model is an hdf5 file which contains the basis and weights of the source model!
try:
self.model = h5py.File(model,w)
self.src_loc = self.model['coordinates']
self.freq = self.model['frequencies']
# Presumably, these arrays are small and will be used very often --> good to have in memory.
self.distr_basis = self.model['distr_basis'][:]
self.spect_basis = self.model['spect_basis'][:]
self.distr_weights = self.model['distr_weights'][:]
# The surface area of each grid element...new since June 18
try:
self.surf_area = self.model['surf_areas'][:]
except KeyError:
# approximate as spherical surface elements...
self.surf_area = get_spherical_surface_elements(
self.src_loc[0],self.src_loc[1])
np.save('surface_areas_grid.npy',self.surf_area)
self.spatial_source_model = self.expand_distr()
except IOError:
msg = 'Unable to open model file '+model
raise IOError(msg)
# the weights matrix is adapted by updates
########################
weights = np.eye(basis_spec.shape[0], num_bases)
########################
# approximate surface
# areas of all elements
# set up in this way.
########################
if config["voronoi_surface_area"]:
from noisi.util.voronoi_surface_area import get_voronoi_surface_area
grd, surf_areas = get_voronoi_surface_area(grd)
else:
surf_areas = get_spherical_surface_elements(grd[0], grd[1])
########################
# Save to an hdf5 file
########################
with h5py.File(os.path.join(sourcepath, 'step_0', 'starting_model.h5'),
'w') as fh:
fh.create_dataset('coordinates', data=grd.astype(np.float64))
fh.create_dataset('frequencies', data=freq.astype(np.float64))
fh.create_dataset('distr_basis', data=basis_geo.astype(np.float64))
# for now: Geographic model can vary freely.
fh.create_dataset('distr_weights', data=weights)
fh.create_dataset('spect_basis', data=basis_spec.astype(np.float64))
fh.create_dataset('surf_areas', data=surf_areas.astype(np.float64))
########################
weights = np.eye(basis_spec.shape[0],num_bases)
########################
# approximate surface
# areas of all elements
# set up in this way.
########################
if config["voronoi_surface_area"]:
from noisi.util.voronoi_surface_area import get_voronoi_surface_area
grd, surf_areas = get_voronoi_surface_area(grd)
else:
surf_areas = get_spherical_surface_elements(grd[0],grd[1])
########################
# Save to an hdf5 file
########################
with h5py.File(os.path.join(sourcepath,'step_0','starting_model.h5'),'w') as fh:
fh.create_dataset('coordinates',data=grd.astype(np.float64))
fh.create_dataset('frequencies',data=freq.astype(np.float64))
fh.create_dataset('distr_basis',data=basis_geo.astype(np.float64))
# for now: Geographic model can vary freely.
fh.create_dataset('distr_weights',data=weights)
fh.create_dataset('spect_basis',data=basis_spec.astype(np.float64))
fh.create_dataset('surf_areas',data=surf_areas.astype(np.float64))
def setup_source_startingmodel(self, args):
# plotting:
colors = ['purple', 'g', 'b', 'orange']
colors_cmaps = [
plt.cm.Purples, plt.cm.Greens, plt.cm.Blues, plt.cm.Oranges
]
print("Setting up source starting model.", end="\n")
with io.open(os.path.join(args.source_model, 'source_config.yml'),
'r') as fh:
source_conf = yaml.safe_load(fh)
with io.open(os.path.join(source_conf['project_path'], 'config.yml'),
'r') as fh:
conf = yaml.safe_load(fh)
with io.open(source_conf['source_setup_file'], 'r') as fh:
parameter_sets = yaml.safe_load(fh)
if conf['verbose']:
print("The following input parameters are used:", end="\n")
pp = pprint.PrettyPrinter()
pp.pprint(parameter_sets)
# load the source locations of the grid
grd = np.load(os.path.join(conf['project_path'], 'sourcegrid.npy'))
# add the approximate spherical surface elements
if grd.shape[-1] < 50000:
surf_el = get_spherical_surface_elements(grd[0], grd[1])
else:
warn('Large grid; surface element computation slow. Using \
approximate surface elements.')
surf_el = np.ones(grd.shape[-1]) * conf['grid_dx_in_m']**2
# get the relevant array sizes
wfs = glob(os.path.join(conf['project_path'], 'greens', '*.h5'))
if wfs != []:
if conf['verbose']:
print('Found wavefield stats.')
else:
pass
else:
raise FileNotFoundError('No wavefield database found. Run \
precompute_wavefield first.')
with WaveField(wfs[0]) as wf:
df = wf.stats['Fs']
n = wf.stats['npad']
freq = np.fft.rfftfreq(n, d=1. / df)
n_distr = len(parameter_sets)
coeffs = np.zeros((grd.shape[-1], n_distr))
spectra = np.zeros((n_distr, len(freq)))
# fill in the distributions and the spectra
for i in range(n_distr):
coeffs[:, i] = self.distribution_from_parameters(
grd, parameter_sets[i], conf['verbose'])
# plot
outfile = os.path.join(args.source_model,
'source_starting_model_distr%g.png' % i)
if create_plot:
plot_grid(grd[0],
grd[1],
coeffs[:, i],
outfile=outfile,
cmap=colors_cmaps[i % len(colors_cmaps)],
sequential=True,
normalize=False,
quant_unit='Spatial weight (-)',
axislabelpad=-0.1,
size=10)
spectra[i, :] = self.spectrum_from_parameters(
freq, parameter_sets[i])
# plotting the spectra
# plotting is not necessarily done to make sure code runs on clusters
if create_plot:
fig1 = plt.figure()
ax = fig1.add_subplot('111')
for i in range(n_distr):
ax.plot(freq,
spectra[i, :] / spectra.max(),
color=colors[i % len(colors_cmaps)])
ax.set_xlabel('Frequency / Nyquist Frequency')
plt.xticks([
0,
freq.max() * 0.25,
freq.max() * 0.5,
freq.max() * 0.75,
freq.max()
], ['0', '0.25', '0.5', '0.75', '1'])
ax.set_ylabel('Rel. PSD norm. to strongest spectrum (-)')
fig1.savefig(
os.path.join(args.source_model,
'source_starting_model_spectra.png'))
# Save to an hdf5 file
with h5py.File(
os.path.join(args.source_model, 'iteration_0',
'starting_model.h5'), 'w') as fh:
fh.create_dataset('coordinates', data=grd)
fh.create_dataset('frequencies', data=freq)
fh.create_dataset('model', data=coeffs.astype(np.float))
fh.create_dataset('spectral_basis', data=spectra.astype(np.float))
fh.create_dataset('surface_areas', data=surf_el.astype(np.float))
# Save to an hdf5 file
with h5py.File(os.path.join(args.source_model, 'spectral_model.h5'),
'w') as fh:
uniform_spatial = np.ones(coeffs.shape) * 1.0
fh.create_dataset('coordinates', data=grd)
fh.create_dataset('frequencies', data=freq)
fh.create_dataset('model', data=uniform_spatial.astype(np.float))
fh.create_dataset('spectral_basis', data=spectra.astype(np.float))
fh.create_dataset('surface_areas', data=surf_el.astype(np.float))