def get_gold_labels(self,
cand_lists: List[List[Candidate]],
annotator: Optional[str] = None) -> List[np.ndarray]:
"""Load dense matrix of GoldLabels for each candidate_class.
:param cand_lists: The candidates to get gold labels for.
:type cand_lists: List of list of candidates.
:param annotator: A specific annotator key to get labels for. Default
None.
:type annotator: str
:raises ValueError: If get_gold_labels is called before gold labels are
loaded, the result will contain ABSTAIN values. We raise a
ValueError to help indicate this potential mistake to the user.
:return: A list of MxN dense matrix where M are the candidates and N is the
annotators. If annotator is provided, return a list of Mx1 matrix.
:rtype: list[np.ndarray]
"""
gold_labels = [
unshift_label_matrix(m) for m in get_sparse_matrix(
self.session, GoldLabelKey, cand_lists, key=annotator)
]
for cand_labels in gold_labels:
if ABSTAIN in cand_labels:
raise ValueError("Gold labels contain ABSTAIN labels. "
"Did you load gold labels beforehand?")
return gold_labels
def convert_labels_to_matrix(labels: List[Dict[str, Any]],
keys: List[str]) -> np.ndarray:
"""Convert labels (the output from LabelerUDF.apply) into a dense matrix.
Note that the input labels are 0-indexed (``{0, 1, ..., k}``),
while the output labels are -1-indexed (``{-1, 0, ..., k-1}``).
:param labels: a list of label mapping (key: key, value=label).
:param keys: a list of all keys.
"""
return unshift_label_matrix(_convert_mappings_to_matrix(labels, keys))
def get_label_matrices(self, cand_lists: List[List[Candidate]]) -> List[np.ndarray]:
"""Load dense matrix of Labels for each candidate_class.
:param cand_lists: The candidates to get labels for.
:return: A list of MxN dense matrix where M are the candidates and N is the
labeling functions.
"""
return [
unshift_label_matrix(m)
for m in get_sparse_matrix(self.session, LabelKey, cand_lists)
]
def _L_matrix(labels: List[Dict[str, Any]],
key_names: List[str]) -> np.ndarray:
"""Convert labels (the output from LabelerUDF.apply) into a dense matrix.
Note that :func:`LabelerUDF.apply` returns a list of list of label mapping,
where the outer list represents candidate_classes, while this method takes a list
of label mapping of each candidate_class.
Also note that the input labels are 0-indexed (``{0, 1, ..., k}``),
while the output labels are -1-indexed (``{-1, 0, ..., k-1}``).
:param labels: a list of label mapping (key: key_name, value=label).
:param key_names: a list of all key_names.
"""
return unshift_label_matrix(_F_matrix(labels, key_names))
def test_shift_label_matrix(caplog):
"""Test the label matrix shifter and unshifter."""
caplog.set_level(logging.INFO)
"""
L is a dense label matrix (ABSTAIN as -1) with values:
-1 0
1 -1
"""
L = np.array([[-1, 0], [1, -1]])
"""
L_sparse is a sparse label matrix (ABSTAIN as 0)
0 1
2 0
"""
L_sparse = shift_label_matrix(L)
assert np.array_equal(L, unshift_label_matrix(L_sparse))
assert L_sparse.count_nonzero() == 2
def get_gold_labels(self,
cand_lists: List[List[Candidate]],
annotator: Optional[str] = None) -> List[np.ndarray]:
"""Load dense matrix of GoldLabels for each candidate_class.
:param cand_lists: The candidates to get gold labels for.
:type cand_lists: List of list of candidates.
:param annotator: A specific annotator key to get labels for. Default
None.
:type annotator: str
:return: A list of MxN dense matrix where M are the candidates and N is the
annotators. If annotator is provided, return a list of Mx1 matrix.
:rtype: list[np.ndarray]
"""
return [
unshift_label_matrix(m) for m in get_sparse_matrix(
self.session, GoldLabelKey, cand_lists, key=annotator)
]
