def _check_single_elem(element):
# Nested array annotations not allowed currently
# If element is a list or a np.ndarray, it's not conform except if it's a quantity of
# length 1
if isinstance(element, list) or (isinstance(element, np.ndarray) and not (
isinstance(element, pq.Quantity) and (
element.shape == () or element.shape == (1,)))):
raise ValueError("Array annotations should only be 1-dimensional")
if isinstance(element, dict):
raise ValueError("Dictionaries are not supported as array annotations")
# Perform regular check for elements of array or list
_check_annotations(element)
def _check_single_elem(element):
# Nested array annotations not allowed currently
# If element is a list or a np.ndarray, it's not conform except if it's a quantity of
# length 1
if isinstance(element, list) or (isinstance(element, np.ndarray) and not (
isinstance(element, pq.Quantity) and (
element.shape == () or element.shape == (1,)))):
raise ValueError("Array annotations should only be 1-dimensional")
if isinstance(element, dict):
raise ValueError("Dictionaries are not supported as array annotations")
# Perform regular check for elements of array or list
_check_annotations(element)
def annotate(self, **annotations):
"""
Add an annotation to the sound
:param annotations: comma-separated key-value pairs to be added to the sound objects annotations. All values
are also stored in the database.
Example:
sound.annotate(type="sentence", transcript="This is a test")
sound.annotations["type"]
sound.annotations["transcript"]
"""
_check_annotations(annotations)
self.annotations.update(annotations)
self.manager.database.store_annotations(self.id, **annotations)
def annotate(self, **annotations):
"""
Add an annotation to the sound
:param annotations: comma-separated key-value pairs to be added to the sound objects annotations. All values
are also stored in the database.
Example:
sound.annotate(type="sentence", transcript="This is a test")
sound.annotations["type"]
sound.annotations["transcript"]
"""
_check_annotations(annotations)
self.annotations.update(annotations)
self.manager.database.store_annotations(self.id, **annotations)
def test__check_annotations__dict(self):
names = ['value%s' % i for i in range(len(self.values))]
values = dict(zip(names, self.values))
_check_annotations(values)
def test__check_annotations__tuple(self):
_check_annotations(tuple(self.values))
_check_annotations((self.values, self.values))
def test__check_annotations__list(self):
_check_annotations(self.values)
def test__check_annotations__valid_dtypes(self):
for value in self.values:
_check_annotations(value)
def _normalize_array_annotations(value, length):
"""Check consistency of array annotations
Recursively check that value is either an array or list containing only "simple" types
(number, string, date/time) or is a dict of those.
Args:
:value: (np.ndarray, list or dict) value to be checked for consistency
:length: (int) required length of the array annotation
Returns:
np.ndarray The array_annotations from value in correct form
Raises:
ValueError: In case value is not accepted as array_annotation(s)
"""
# First stage, resolve dict of annotations into single annotations
if isinstance(value, dict):
for key in value.keys():
if isinstance(value[key], dict):
raise ValueError(
"Nested dicts are not allowed as array annotations")
value[key] = _normalize_array_annotations(value[key], length)
elif value is None:
raise ValueError("Array annotations must not be None")
# If not array annotation, pass on to regular check and make it a list, that is checked again
# This covers array annotations with length 1
elif not isinstance(value,
(list, np.ndarray)) or (isinstance(value, pq.Quantity)
and value.shape == ()):
_check_annotations(value)
value = _normalize_array_annotations(np.array([value]), length)
# If array annotation, check for correct length, only single dimension and allowed data
else:
# Get length that is required for array annotations, which is equal to the length
# of the object's data
own_length = length
# Escape check if empty array or list and just annotate an empty array (length 0)
# This enables the user to easily create dummy array annotations that will be filled
# with data later on
if len(value) == 0:
if not isinstance(value, np.ndarray):
value = np.ndarray((0, ))
val_length = own_length
else:
# Note: len(o) also works for np.ndarray, it then uses the first dimension,
# which is exactly the desired behaviour here
val_length = len(value)
if not own_length == val_length:
raise ValueError(
"Incorrect length of array annotation: {} != {}".format(
val_length, own_length))
# Local function used to check single elements of a list or an array
# They must not be lists or arrays and fit the usual annotation data types
def _check_single_elem(element):
# Nested array annotations not allowed currently
# If element is a list or a np.ndarray, it's not conform except if it's a quantity of
# length 1
if isinstance(element, list) or (
isinstance(element, np.ndarray)
and not (isinstance(element, pq.Quantity) and
(element.shape == () or element.shape == (1, )))):
raise ValueError(
"Array annotations should only be 1-dimensional")
if isinstance(element, dict):
raise ValueError(
"Dictionaries are not supported as array annotations")
# Perform regular check for elements of array or list
_check_annotations(element)
# Arrays only need testing of single element to make sure the others are the same
if isinstance(value, np.ndarray):
# Type of first element is representative for all others
# Thus just performing a check on the first element is enough
# Even if it's a pq.Quantity, which can be scalar or array, this is still true
# Because a np.ndarray cannot contain scalars and sequences simultaneously
# If length of data is 0, then nothing needs to be checked
if len(value):
# Perform check on first element
_check_single_elem(value[0])
return value
# In case of list, it needs to be ensured that all data are of the same type
else:
# Conversion to numpy array makes all elements same type
# Converts elements to most general type
try:
value = np.array(value)
# Except when scalar and non-scalar values are mixed, this causes conversion to fail
except ValueError as e:
msg = str(e)
if "setting an array element with a sequence." in msg:
raise ValueError(
"Scalar values and arrays/lists cannot be "
"combined into a single array annotation")
else:
raise e
# If most specialized data type that possibly fits all elements is object,
# raise an Error with a telling error message, because this means the elements
# are not compatible
if value.dtype == object:
raise ValueError(
"Cannot convert list of incompatible types into a single"
" array annotation")
# Check the first element for correctness
# If its type is correct for annotations, all others are correct as well
# Note: Emtpy lists cannot reach this point
_check_single_elem(value[0])
return value
def test__check_annotations__dict(self):
names = ['value%s' % i for i in range(len(self.values))]
values = dict(zip(names, self.values))
_check_annotations(values)
def test__check_annotations__tuple(self):
_check_annotations(tuple(self.values))
_check_annotations((self.values, self.values))
def test__check_annotations__list(self):
_check_annotations(self.values)
def test__check_annotations__valid_dtypes(self):
for value in self.values:
_check_annotations(value)
def _normalize_array_annotations(value, length):
"""Check consistency of array annotations
Recursively check that value is either an array or list containing only "simple" types
(number, string, date/time) or is a dict of those.
Args:
:value: (np.ndarray, list or dict) value to be checked for consistency
:length: (int) required length of the array annotation
Returns:
np.ndarray The array_annotations from value in correct form
Raises:
ValueError: In case value is not accepted as array_annotation(s)
"""
# First stage, resolve dict of annotations into single annotations
if isinstance(value, dict):
for key in value.keys():
if isinstance(value[key], dict):
raise ValueError("Nested dicts are not allowed as array annotations")
value[key] = _normalize_array_annotations(value[key], length)
elif value is None:
raise ValueError("Array annotations must not be None")
# If not array annotation, pass on to regular check and make it a list, that is checked again
# This covers array annotations with length 1
elif not isinstance(value, (list, np.ndarray)) or (
isinstance(value, pq.Quantity) and value.shape == ()):
_check_annotations(value)
value = _normalize_array_annotations(np.array([value]), length)
# If array annotation, check for correct length, only single dimension and allowed data
else:
# Get length that is required for array annotations, which is equal to the length
# of the object's data
own_length = length
# Escape check if empty array or list and just annotate an empty array (length 0)
# This enables the user to easily create dummy array annotations that will be filled
# with data later on
if len(value) == 0:
if not isinstance(value, np.ndarray):
value = np.ndarray((0,))
val_length = own_length
else:
# Note: len(o) also works for np.ndarray, it then uses the first dimension,
# which is exactly the desired behaviour here
val_length = len(value)
if not own_length == val_length:
raise ValueError(
"Incorrect length of array annotation: {} != {}".format(val_length, own_length))
# Local function used to check single elements of a list or an array
# They must not be lists or arrays and fit the usual annotation data types
def _check_single_elem(element):
# Nested array annotations not allowed currently
# If element is a list or a np.ndarray, it's not conform except if it's a quantity of
# length 1
if isinstance(element, list) or (isinstance(element, np.ndarray) and not (
isinstance(element, pq.Quantity) and (
element.shape == () or element.shape == (1,)))):
raise ValueError("Array annotations should only be 1-dimensional")
if isinstance(element, dict):
raise ValueError("Dictionaries are not supported as array annotations")
# Perform regular check for elements of array or list
_check_annotations(element)
# Arrays only need testing of single element to make sure the others are the same
if isinstance(value, np.ndarray):
# Type of first element is representative for all others
# Thus just performing a check on the first element is enough
# Even if it's a pq.Quantity, which can be scalar or array, this is still true
# Because a np.ndarray cannot contain scalars and sequences simultaneously
# If length of data is 0, then nothing needs to be checked
if len(value):
# Perform check on first element
_check_single_elem(value[0])
return value
# In case of list, it needs to be ensured that all data are of the same type
else:
# Conversion to numpy array makes all elements same type
# Converts elements to most general type
try:
value = np.array(value)
# Except when scalar and non-scalar values are mixed, this causes conversion to fail
except ValueError as e:
msg = str(e)
if "setting an array element with a sequence." in msg:
raise ValueError("Scalar values and arrays/lists cannot be "
"combined into a single array annotation")
else:
raise e
# If most specialized data type that possibly fits all elements is object,
# raise an Error with a telling error message, because this means the elements
# are not compatible
if value.dtype == object:
raise ValueError("Cannot convert list of incompatible types into a single"
" array annotation")
# Check the first element for correctness
# If its type is correct for annotations, all others are correct as well
# Note: Emtpy lists cannot reach this point
_check_single_elem(value[0])
return value