def get_fields(self):
""" Get a list of fields that all inputs have
Returns:
list(verif.field.Field): A list of fields
"""
all_fields = set()
for f in range(self._get_num_inputs_with_clim()):
input = self._inputs[f]
if f == 0:
all_fields = input.get_fields()
else:
all_fields = set(all_fields) & set(input.get_fields())
return list(all_fields)
def test_valid2(self):
input = verif.input.Netcdf("verif/tests/files/netcdf_valid2.nc")
locations = input.locations
self.assertTrue(1, len(locations))
self.assertEqual(verif.location.Location(18700, 59.9423, 10.72, 94), locations[0])
np.testing.assert_array_equal(np.array([0, 1, 2]), input.leadtimes)
np.testing.assert_array_equal(np.array([1388534400, 1388620800]), input.times)
obs = input.obs
self.assertEqual(2, obs.shape[0]) # Time
self.assertEqual(3, obs.shape[1]) # Leadtime
self.assertEqual(1, obs.shape[2]) # Location
self.assertAlmostEqual(1, obs[0, 0, 0])
self.assertAlmostEqual(2, obs[0, 1, 0])
self.assertAlmostEqual(3, obs[0, 2, 0])
self.assertAlmostEqual(4, obs[1, 0, 0])
self.assertAlmostEqual(5, obs[1, 1, 0])
self.assertAlmostEqual(6, obs[1, 2, 0])
self.assertEqual(0, len(input.thresholds))
self.assertEqual(0, len(input.quantiles))
self.assertTrue(input.ensemble is None)
self.assertTrue(verif.field.Obs() in input.get_fields())
self.assertFalse(verif.field.Fcst() in input.get_fields())
def main():
parser = argparse.ArgumentParser(
prog="text2verif",
description="Convert between Verif text and NetCDF files")
parser.add_argument('ifile', type=str, help="Verif text file (input)")
parser.add_argument('ofile', type=str, help="Verif NetCDF file (output)")
parser.add_argument('--debug',
help='Print debug information',
action="store_true")
if len(sys.argv) == 1:
parser.print_help()
sys.exit(0)
args = parser.parse_args()
input = verif.input.get_input(args.ifile)
times = input.times
leadtimes = input.leadtimes
locations = input.locations
variable = input.variable
output = netCDF4.Dataset(args.ofile, 'w')
output.createDimension("time", None)
output.createDimension("leadtime", len(leadtimes))
output.createDimension("location", len(locations))
thresholds = input.thresholds
quantiles = input.quantiles
if len(thresholds) > 0:
if args.debug:
print("Adding %d thresholds" % len(thresholds))
output.createDimension("threshold", len(thresholds))
output.createVariable("threshold", "f4", ["threshold"])
output["threshold"][:] = thresholds
output.createVariable("cdf", "f4",
("time", "leadtime", "location", "threshold"))
output.variables["cdf"][:] = input.threshold_scores
if len(quantiles) > 0:
if args.debug:
print("Adding %d quantiles" % len(quantiles))
output.createDimension("quantile", len(quantiles))
output.createVariable("quantile", "f4", ["quantile"])
output["quantile"][:] = quantiles
output.createVariable("x", "f4",
("time", "leadtime", "location", "quantile"))
output.variables["x"][:] = input.quantile_scores
vTime = output.createVariable("time", "i4", ("time", ))
vOffset = output.createVariable("leadtime", "f4", ("leadtime", ))
vLocation = output.createVariable("location", "i4", ("location", ))
vLat = output.createVariable("lat", "f4", ("location", ))
vLon = output.createVariable("lon", "f4", ("location", ))
vElev = output.createVariable("altitude", "f4", ("location", ))
vfcst = output.createVariable("fcst", "f4",
("time", "leadtime", "location"))
vobs = output.createVariable("obs", "f4", ("time", "leadtime", "location"))
# Create nonstandard fields
standard = [verif.field.Obs(), verif.field.Fcst()]
fields = [field for field in input.get_fields() if field not in standard]
for field in fields:
name = field.name()
if field.__class__ == verif.field.Other:
if args.debug:
print("Adding non-standard score '%s'" % name)
output.createVariable(name, "f4", ("time", "leadtime", "location"))
output.variables[name][:] = input.other_score(field.name())[:]
output.standard_name = variable.name
output.units = unit = variable.units.replace("$", "")
vobs[:] = input.obs
vfcst[:] = input.fcst
vTime[:] = input.times
vOffset[:] = input.leadtimes
vLocation[:] = [s.id for s in locations]
vLat[:] = [s.lat for s in locations]
vLon[:] = [s.lon for s in locations]
vElev[:] = [s.elev for s in locations]
output.Conventions = "verif_1.0.0"
output.close()
def _get_score(self, field, input_index):
""" Load the field variable from input, but only include the common data
Scores loaded will have the same dimension, regardless what input_index
is used.
field: The type is of verif.field
input_index: which input to load from
"""
num_inputs = self._get_num_inputs_with_clim()
if field == verif.field.Obs():
"""
Treat observation different. Since the observations should be the same
for all inputs, reuse the observations if one or more inputs are
missing these
"""
field = self._obs_field
if field in self._get_score_cache[input_index]:
return self._get_score_cache[input_index][field]
found_obs = False
for i in range(num_inputs):
input = self._inputs[i]
all_fields = input.get_fields()
temp = input.obs
if field in all_fields:
Itimes = self._get_time_indices(i)
Ileadtimes = self._get_leadtime_indices(i)
Ilocations = self._get_location_indices(i)
temp = temp[Itimes, :, :]
temp = temp[:, Ileadtimes, :]
temp = temp[:, :, Ilocations]
self._get_score_cache[i][field] = temp
found_obs = True
if not found_obs:
verif.util.error("No files have observations")
for i in range(num_inputs):
if field not in self._get_score_cache[i]:
for j in range(num_inputs):
if field in self._get_score_cache[j]:
verif.util.warning(
"No observations in %s. Loading from %s" %
(self._inputs[i].fullname,
self._inputs[j].fullname))
self._get_score_cache[i][
field] = self._get_score_cache[j][field]
break
else:
# Check if data is cached
if field in self._get_score_cache[input_index]:
return self._get_score_cache[input_index][field]
if field == verif.field.Fcst():
field = self._fcst_field
for i in range(num_inputs):
if field not in self._get_score_cache[i]:
input = self._inputs[i]
all_fields = input.get_fields()
if field not in all_fields:
verif.util.error("%s does not contain '%s'" %
(self.get_names()[i], field.name()))
elif field == verif.field.Obs():
temp = input.obs
elif field == verif.field.Fcst():
temp = input.fcst
elif field == verif.field.Pit():
temp = input.pit
x0 = self.variable.x0
x1 = self.variable.x1
if x0 is not None or x1 is not None:
temp = verif.field.Pit.randomize(
input.obs, temp, x0, x1)
elif field.__class__ is verif.field.Ensemble:
temp = input.ensemble[:, :, :, field.member]
elif field.__class__ is verif.field.Threshold:
I = np.where(input.thresholds == field.threshold)[0]
assert (len(I) == 1)
temp = input.threshold_scores[:, :, :, I[0]]
elif field.__class__ is verif.field.Quantile:
I = np.where(input.quantiles == field.quantile)[0]
assert (len(I) == 1)
temp = input.quantile_scores[:, :, :, I[0]]
else:
temp = input.other_score(field.name())
Itimes = self._get_time_indices(i)
Ileadtimes = self._get_leadtime_indices(i)
Ilocations = self._get_location_indices(i)
temp = temp[Itimes, :, :]
temp = temp[:, Ileadtimes, :]
temp = temp[:, :, Ilocations]
self._get_score_cache[i][field] = temp
"""
Remove missing. If one configuration has a missing value, set all
configurations to missing. This can happen when the times are
available, but have missing values.
"""
if self._remove_missing_across_all:
is_missing = np.isnan(self._get_score_cache[0][field])
for i in range(1, num_inputs):
is_missing = is_missing | (np.isnan(
self._get_score_cache[i][field]))
for i in range(num_inputs):
self._get_score_cache[i][field][is_missing] = np.nan
return self._get_score_cache[input_index][field]
def _get_score(self, field, input_index):
""" Load the field variable from input, but only include the common data
Scores loaded will have the same dimension, regardless what input_index
is used.
field: The type is of verif.field
input_index: which input to load from
"""
# Check if data is cached
if (field in self._get_score_cache[input_index]):
return self._get_score_cache[input_index][field]
if field == verif.field.Obs():
field = self._obs_field
if field == verif.field.Fcst():
field = self._fcst_field
# Load all inputs
for i in range(0, self._get_num_inputs_with_clim()):
if (field not in self._get_score_cache[i]):
input = self._inputs[i]
all_fields = input.get_fields() + [
verif.field.ObsWindow(),
verif.field.FcstWindow()
]
if (field not in all_fields):
verif.util.error("%s does not contain '%s'" %
(self.get_names()[i], field.name()))
if field == verif.field.Obs():
temp = input.obs
elif field == verif.field.Fcst():
temp = input.fcst
elif field == verif.field.Pit():
temp = input.pit
x0 = self.variable.x0
x1 = self.variable.x1
if x0 is not None or x1 is not None:
# w = ""
# if x0 is not None:
# w += " obs=%g" % x0
# if x1 is not None:
# w += " obs=%g" % x1
# verif.util.warning("Randomizing PIT values where %s" + w)
temp = verif.field.Pit.randomize(
input.obs, temp, x0, x1)
elif field.__class__ is verif.field.Ensemble:
temp = input.ensemble[:, :, :, field.member]
elif field.__class__ is verif.field.Threshold:
I = np.where(input.thresholds == field.threshold)[0]
assert (len(I) == 1)
temp = input.threshold_scores[:, :, :, I[0]]
elif field.__class__ is verif.field.Quantile:
I = np.where(input.quantiles == field.quantile)[0]
assert (len(I) == 1)
temp = input.quantile_scores[:, :, :, I[0]]
elif field == verif.field.ObsWindow():
temp = input.obs[:, :, :]
temp = self._calculate_window(temp, input.leadtimes)
elif field == verif.field.FcstWindow():
temp = input.fcst[:, :, :]
temp = self._calculate_window(temp, input.leadtimes)
else:
temp = input.other_score(field.name())
Itimes = self._get_time_indices(i)
Ileadtimes = self._get_leadtime_indices(i)
Ilocations = self._get_location_indices(i)
temp = temp[Itimes, :, :]
temp = temp[:, Ileadtimes, :]
temp = temp[:, :, Ilocations]
self._get_score_cache[i][field] = temp
# Remove missing. If one configuration has a missing value, set all
# configurations to missing. This can happen when the times are
# available, but have missing values.
if self._remove_missing_across_all:
is_missing = np.isnan(self._get_score_cache[0][field])
for i in range(1, self._get_num_inputs_with_clim()):
is_missing = is_missing | (np.isnan(
self._get_score_cache[i][field]))
for i in range(0, self._get_num_inputs_with_clim()):
self._get_score_cache[i][field][is_missing] = np.nan
return self._get_score_cache[input_index][field]
