def get_numpy_scores_and_labels_filtered_by_labels(self, pai_labels=None):
if not pai_labels:
return self.get_numpy_scores(), self.get_numpy_labels()
ids = self.scores.keys()
annotations_from_ids = annotations.get_annotations_from_ids(ids)
# filtered_ids = self._get_filtered_ids(
# annotations_from_ids, Filter(spai=Spai.GENUINE)
# )
scores = []
labels = []
for id in ids:
annotation = [
annotation for annotation in annotations_from_ids if annotation.id == id
][0]
if annotation.categorization.get("fine_grained_pai") == 0:
scores.append(self.scores[id])
labels.append(0)
elif annotation.categorization.get("fine_grained_pai") in pai_labels:
scores.append(self.scores[id])
labels.append(1)
return (
np.asarray(list(scores), dtype=np.float32),
np.asarray(labels, dtype=np.int),
)
def filtered_by(self, filter: Filter):
ids = self.scores.keys()
annotations_from_ids = annotations.get_annotations_from_ids(ids)
filtered_ids = self._get_filtered_ids(annotations_from_ids, filter)
return {key: value for key, value in self.scores.items() if key in filtered_ids}
def get_numpy_labels(self):
ids = self.scores.keys()
annotations_from_ids = annotations.get_annotations_from_ids(ids)
filtered_ids = self._get_filtered_ids(
annotations_from_ids, Filter(scenario=Scenario.GENUINE)
)
labels = [0 if id in filtered_ids else 1 for id in ids]
return np.asarray(labels, dtype=np.int)
def get_attacks(self):
ids = self.scores.keys()
annotations_from_ids = annotations.get_annotations_from_ids(ids)
filtered_ids = self._get_filtered_ids(
annotations_from_ids, Filter(scenario=Scenario.GENUINE)
)
scores = [score for id, score in self.scores.items() if id not in filtered_ids]
return scores
def get_numpy_fine_grained_pai_labels(self):
ids = self.scores.keys()
annotations_from_ids = annotations.get_annotations_from_ids(ids)
fine_grained_pai_labels = [
annotation.categorization.get("fine_grained_pai")
for annotation in annotations_from_ids
]
return np.asarray(fine_grained_pai_labels, dtype=np.int)
def _get_numpy_labels_filter_by_filter(
self,
options: List,
filter_provider: Callable,
unknown_label_value: int = None,
encode_label: bool = False,
):
ids = self.scores.keys()
annotations_from_ids = annotations.get_annotations_from_ids(ids)
filtered_ids_by_subdivision = {}
filtered_ids_order = {}
for order, option in enumerate(options):
filtered_ids_by_subdivision[option] = self._get_filtered_ids(
annotations_from_ids, filter_provider(option)
)
filtered_ids_order[option] = order
labels = []
for id in ids:
value = None
for subdivision, filtered_ids in filtered_ids_by_subdivision.items():
if id in filtered_ids:
value = subdivision.value
break
if value is None and unknown_label_value is not None:
value = unknown_label_value
labels.append(value)
if encode_label:
le = LabelEncoder()
labels = le.fit_transform(labels)
le_name_mapping = dict(zip(le.classes_, le.transform(le.classes_)))
if le_name_mapping.get(str(unknown_label_value)) is not None:
labels = [label - 1 for label in labels]
return np.asarray(labels, dtype=np.int)