def test_complex_case(self):
expr = (
'(not quality_flags.invalid'
' and not pixel_classif_flags.IDEPIX_CLOUD'
' and not pixel_classif_flags.IDEPIX_CLOUD_BUFFER'
' and not pixel_classif_flags.IDEPIX_CLOUD_SHADOW'
' and not pixel_classif_flags.IDEPIX_SNOW_ICE'
' and not (c2rcc_flags.Rtosa_OOS and conc_chl > 1.0)'
' and not c2rcc_flags.Rtosa_OOR'
' and not c2rcc_flags.Rhow_OOR'
' and not (c2rcc_flags.Cloud_risk and immersed_cyanobacteria == 0)'
' and floating_vegetation == 0'
' and conc_chl > 0.01'
' and not (floating_cyanobacteria == 1 or chl_pitarch > 500))')
quality_flags = namedtuple('quality_flags', ['invalid'])
quality_flags.invalid = np.array([0])
pixel_classif_flags = namedtuple('pixel_classif_flags', [
'IDEPIX_CLOUD', 'IDEPIX_CLOUD_BUFFER', 'IDEPIX_CLOUD_SHADOW',
'IDEPIX_SNOW_ICE'
])
pixel_classif_flags.IDEPIX_CLOUD = np.array([0])
pixel_classif_flags.IDEPIX_CLOUD_BUFFER = np.array([0])
pixel_classif_flags.IDEPIX_CLOUD_SHADOW = np.array([0])
pixel_classif_flags.IDEPIX_SNOW_ICE = np.array([0])
c2rcc_flags = namedtuple(
'c2rcc_flags',
['Rtosa_OOS', 'Rtosa_OOR', 'Rhow_OOR', 'Cloud_risk'])
c2rcc_flags.Rtosa_OOS = np.array([0])
c2rcc_flags.Rtosa_OOR = np.array([0])
c2rcc_flags.Rhow_OOR = np.array([0])
c2rcc_flags.Cloud_risk = np.array([0])
namespace = dict(
np=np,
quality_flags=quality_flags,
pixel_classif_flags=pixel_classif_flags,
c2rcc_flags=c2rcc_flags,
immersed_cyanobacteria=np.array([0]),
floating_cyanobacteria=np.array([0]),
floating_vegetation=np.array([0]),
conc_chl=np.array([0]),
chl_pitarch=np.array([0]),
)
actual_value = compute_array_expr(expr, namespace=namespace)
expected_value = 0
npt.assert_array_almost_equal(actual_value, np.array([expected_value]))
namespace['conc_chl'] = np.array([0.2])
actual_value = compute_array_expr(expr, namespace=namespace)
expected_value = 1
npt.assert_array_almost_equal(actual_value, np.array([expected_value]))
pixel_classif_flags.IDEPIX_CLOUD_SHADOW = np.array([0.2])
actual_value = compute_array_expr(expr, namespace=namespace)
expected_value = 0
npt.assert_array_almost_equal(actual_value, np.array([expected_value]))
def test_valid_exprs(self):
namespace = dict(a=np.array([0.1, 0.3, 0.1, 0.7, 0.4, 0.9]),
b=np.array([0.2, 0.1, 0.3, 0.2, 0.4, 0.8]),
np=np,
xr=xr)
value = compute_array_expr('a + 1', namespace=namespace)
np.testing.assert_array_almost_equal(value,
np.array([1.1, 1.3, 1.1, 1.7, 1.4, 1.9]))
value = compute_array_expr('a * b', namespace=namespace)
np.testing.assert_array_almost_equal(value,
np.array([0.02, 0.03, 0.03, 0.14, 0.16, 0.72]))
value = compute_array_expr('max(a, b)', namespace=namespace)
np.testing.assert_array_almost_equal(value,
np.array([0.2, 0.3, 0.3, 0.7, 0.4, 0.9]))
value = compute_array_expr('a > b', namespace=namespace)
np.testing.assert_equal(value,
np.array([False, True, False, True, False, True]))
value = compute_array_expr('a == b', namespace=namespace)
np.testing.assert_equal(value,
np.array([False, False, False, False, True, False]))
# This weirdo expression is a result of translating SNAP conditional expressions to Python.
value = compute_array_expr('a > 0.35 if a else b', namespace=namespace)
np.testing.assert_equal(value,
np.array([0.2, 0.1, 0.3, 0.7, 0.4, 0.9]))
# We actually mean
value = compute_array_expr('where(a > 0.35, a, b)', namespace=namespace)
np.testing.assert_equal(value,
np.array([0.2, 0.1, 0.3, 0.7, 0.4, 0.9]))
def test_valid_exprs(self):
namespace = dict(a=np.array([0.1, 0.3, 0.1, 0.7, 0.4, 0.9]),
b=np.array([0.2, 0.1, 0.3, 0.2, 0.4, 0.8]),
np=np)
value = compute_array_expr('a + 1', namespace=namespace)
np.testing.assert_array_almost_equal(value,
np.array([1.1, 1.3, 1.1, 1.7, 1.4, 1.9]))
value = compute_array_expr('a * b', namespace=namespace)
np.testing.assert_array_almost_equal(value,
np.array([0.02, 0.03, 0.03, 0.14, 0.16, 0.72]))
value = compute_array_expr('max(a, b)', namespace=namespace)
np.testing.assert_array_almost_equal(value,
np.array([0.2, 0.3, 0.3, 0.7, 0.4, 0.9]))
value = compute_array_expr('a > b', namespace=namespace)
np.testing.assert_equal(value,
np.array([False, True, False, True, False, True]))
value = compute_array_expr('a == b', namespace=namespace)
np.testing.assert_equal(value,
np.array([False, False, False, False, True, False]))
def compute_dataset(dataset: xr.Dataset,
processed_variables: NameDictPairList = None,
errors: str = 'raise') -> xr.Dataset:
"""
Compute a dataset from another dataset and return it.
New variables are computed according to the value of an ``expression`` attribute which, if given,
must by a valid Python expression that can reference any other preceding variables by name.
The expression can also reference any flags defined by another variable according the their CF
attributes ``flag_meaning`` and ``flag_values``.
Invalid values may be masked out using the value of an
optional ``valid_pixel_expression`` attribute that forms a boolean Python expression.
The value of the ``_FillValue`` attribute or NaN will be used in the new variable where the
expression returns zero or false.
Other attributes will be stored as variable metadata as-is.
:param dataset: A dataset.
:param processed_variables: Optional list of variables that will be loaded or computed in the order given.
Each variable is either identified by name or by a name to variable attributes mapping.
:param errors: How to deal with errors while evaluating expressions.
May be be one of "raise", "warn", or "ignore".
:return: new dataset with computed variables
"""
if processed_variables:
processed_variables = to_resolved_name_dict_pairs(processed_variables,
dataset,
keep=True)
else:
var_names = list(dataset.data_vars)
var_names = sorted(var_names,
key=functools.partial(_get_var_sort_key, dataset))
processed_variables = [(var_name, None) for var_name in var_names]
# Initialize namespace with some constants and modules
namespace = dict(NaN=np.nan, PI=math.pi, np=np, xr=xr)
# Now add all mask sets and variables
for var_name in dataset.data_vars:
var = dataset[var_name]
if MaskSet.is_flag_var(var):
namespace[var_name] = MaskSet(var)
else:
namespace[var_name] = var
for var_name, var_props in processed_variables:
if var_name in dataset.data_vars:
# Existing variable
var = dataset[var_name]
if var_props:
var_props_temp = var_props
var_props = dict(var.attrs)
var_props.update(var_props_temp)
else:
var_props = dict(var.attrs)
else:
# Computed variable
var = None
if var_props is None:
var_props = dict()
expression = var_props.get('expression')
if expression:
# Compute new variable
computed_array = compute_array_expr(expression,
namespace=namespace,
result_name=f'{var_name!r}',
errors=errors)
if computed_array is not None:
if hasattr(computed_array, 'attrs'):
var = computed_array
var.attrs.update(var_props)
namespace[var_name] = computed_array
valid_pixel_expression = var_props.get('valid_pixel_expression')
if valid_pixel_expression:
# Compute new mask for existing variable
if var is None:
raise ValueError(f'undefined variable {var_name!r}')
valid_mask = compute_array_expr(
valid_pixel_expression,
namespace=namespace,
result_name=f'valid mask for {var_name!r}',
errors=errors)
if valid_mask is not None:
masked_var = var.where(valid_mask)
if hasattr(masked_var, 'attrs'):
masked_var.attrs.update(var_props)
namespace[var_name] = masked_var
computed_dataset = dataset.copy()
for name, value in namespace.items():
if isinstance(value, xr.DataArray):
computed_dataset[name] = value
return computed_dataset
def evaluate_dataset(dataset: xr.Dataset,
processed_variables: NameDictPairList = None,
errors: str = 'raise') -> xr.Dataset:
"""
Compute new variables or mask existing variables in *dataset*
by the evaluation of Python expressions, that may refer to other
existing or new variables.
Returns a new dataset that contains the old and new variables,
where both may bew now masked.
Expressions may be given by attributes of existing variables in
*dataset* or passed a via the *processed_variables* argument
which is a sequence of variable name / attributes tuples.
Two types of expression attributes are recognized in the attributes:
1. The attribute ``expression`` generates
a new variable computed from its attribute value.
2. The attribute ``valid_pixel_expression`` masks out
invalid variable values.
In both cases the attribuite value must be a string that forms
a valid Python expression that can reference any other preceding
variables by name.
The expression can also reference any flags defined by another
variable according the their CF attributes ``flag_meaning``
and ``flag_values``.
Invalid variable values may be masked out using the value the
``valid_pixel_expression`` attribute whose value should form
a Boolean Python expression. In case, the expression
returns zero or false, the value of the ``_FillValue`` attribute
or NaN will be used in the new variable.
Other attributes will be stored as variable metadata as-is.
:param dataset: A dataset.
:param processed_variables: Optional list of variable
name-attributes pairs that will processed in the given order.
:param errors: How to deal with errors while evaluating expressions.
May be be one of "raise", "warn", or "ignore".
:return: new dataset with computed variables
"""
if processed_variables:
processed_variables = to_resolved_name_dict_pairs(processed_variables,
dataset,
keep=True)
else:
var_names = list(dataset.data_vars)
var_names = sorted(var_names,
key=functools.partial(_get_var_sort_key, dataset))
processed_variables = [(var_name, None) for var_name in var_names]
# Initialize namespace with some constants and modules
namespace = dict(NaN=np.nan, PI=math.pi, np=np, xr=xr)
# Now add all mask sets and variables
for var_name in dataset.data_vars:
var = dataset[var_name]
if MaskSet.is_flag_var(var):
namespace[var_name] = MaskSet(var)
else:
namespace[var_name] = var
for var_name, var_props in processed_variables:
if var_name in dataset.data_vars:
# Existing variable
var = dataset[var_name]
if var_props:
var_props_temp = var_props
var_props = dict(var.attrs)
var_props.update(var_props_temp)
else:
var_props = dict(var.attrs)
else:
# Computed variable
var = None
if var_props is None:
var_props = dict()
do_load = var_props.get('load', False)
expression = var_props.get('expression')
if expression:
# Compute new variable
computed_array = compute_array_expr(expression,
namespace=namespace,
result_name=f'{var_name!r}',
errors=errors)
if computed_array is not None:
if hasattr(computed_array, 'attrs'):
var = computed_array
var.attrs.update(var_props)
if do_load:
computed_array.load()
namespace[var_name] = computed_array
valid_pixel_expression = var_props.get('valid_pixel_expression')
if valid_pixel_expression:
# Compute new mask for existing variable
if var is None:
raise ValueError(f'undefined variable {var_name!r}')
valid_mask = compute_array_expr(
valid_pixel_expression,
namespace=namespace,
result_name=f'valid mask for {var_name!r}',
errors=errors)
if valid_mask is not None:
masked_var = var.where(valid_mask)
if hasattr(masked_var, 'attrs'):
masked_var.attrs.update(var_props)
if do_load:
masked_var.load()
namespace[var_name] = masked_var
computed_dataset = dataset.copy()
for name, value in namespace.items():
if isinstance(value, xr.DataArray):
computed_dataset[name] = value
return computed_dataset