def execute(self, slot, subindex, roi, result):
all_features_and_labels_df = None
for lane_index, (labels_dict_slot, features_slot) in \
enumerate( zip(self.EdgeLabelsDict, self.EdgeFeaturesDataFrame) ):
logger.info(
"Retrieving features for lane {}...".format(lane_index))
labels_dict = labels_dict_slot.value.copy(
) # Copy now to avoid threading issues.
if not labels_dict:
continue
sp_columns = np.array(labels_dict.keys())
edge_features_df = features_slot.value
assert list(edge_features_df.columns[0:2]) == ['sp1', 'sp2']
labels_df = pd.DataFrame(sp_columns, columns=['sp1', 'sp2'])
labels_df['label'] = labels_dict.values()
# Drop zero labels
labels_df = labels_df[labels_df['label'] != 0]
# Merge in features
features_and_labels_df = pd.merge(edge_features_df,
labels_df,
how='right',
on=['sp1', 'sp2'])
if all_features_and_labels_df is not None:
all_features_and_labels_df = all_features_and_labels_df.append(
features_and_labels_df)
else:
all_features_and_labels_df = features_and_labels_df
if all_features_and_labels_df is None:
# No labels yet.
result[0] = None
return
assert list(all_features_and_labels_df.columns[0:2]) == ['sp1', 'sp2']
assert all_features_and_labels_df.columns[-1] == 'label'
feature_matrix = all_features_and_labels_df.iloc[:, 2:
-1].values # Omit 'sp1', 'sp2', and 'label'
labels = all_features_and_labels_df.iloc[:, -1].values
logger.info("Training classifier with {} labels...".format(
len(labels)))
# TODO: Allow factory to be configured via an input slot
classifier_factory = ParallelVigraRfLazyflowClassifierFactory()
classifier = classifier_factory.create_and_train(
feature_matrix,
labels,
feature_names=all_features_and_labels_df.columns[2:-1].values)
assert set(classifier.known_classes).issubset(set([1, 2]))
result[0] = classifier
def test_basic(self):
# Initialize factory
factory = ParallelVigraRfLazyflowClassifierFactory(10)
# Train
classifier = factory.create_and_train(self.training_feature_matrix, self.training_labels)
assert isinstance(classifier, ParallelVigraRfLazyflowClassifier)
assert list(classifier.known_classes) == [1,2]
# Predict
probabilities = classifier.predict_probabilities( self.prediction_data )
assert probabilities.shape == (4,2)
assert probabilities.dtype == numpy.float32
assert (0 <= probabilities).all() and (probabilities <= 1.0).all()
assert (numpy.argmax(probabilities, axis=-1)+1 == self.expected_classes).all()
def execute(self, slot, subindex, roi, result):
all_features_and_labels_df = None
for lane_index, (labels_dict_slot, features_slot) in \
enumerate( zip(self.EdgeLabelsDict, self.EdgeFeaturesDataFrame) ):
logger.info("Retrieving features for lane {}...".format(lane_index))
labels_dict = labels_dict_slot.value.copy() # Copy now to avoid threading issues.
if not labels_dict:
continue
sp_columns = np.array(labels_dict.keys())
edge_features_df = features_slot.value
assert list(edge_features_df.columns[0:2]) == ['sp1', 'sp2']
labels_df = pd.DataFrame(sp_columns, columns=['sp1', 'sp2'])
labels_df['label'] = labels_dict.values()
# Drop zero labels
labels_df = labels_df[labels_df['label'] != 0]
# Merge in features
features_and_labels_df = pd.merge(edge_features_df, labels_df, how='right', on=['sp1', 'sp2'])
if all_features_and_labels_df is not None:
all_features_and_labels_df = all_features_and_labels_df.append(features_and_labels_df)
else:
all_features_and_labels_df = features_and_labels_df
if all_features_and_labels_df is None:
# No labels yet.
result[0] = None
return
assert list(all_features_and_labels_df.columns[0:2]) == ['sp1', 'sp2']
assert all_features_and_labels_df.columns[-1] == 'label'
feature_matrix = all_features_and_labels_df.iloc[:, 2:-1].values # Omit 'sp1', 'sp2', and 'label'
labels = all_features_and_labels_df.iloc[:, -1].values
logger.info("Training classifier with {} labels...".format( len(labels) ))
# TODO: Allow factory to be configured via an input slot
classifier_factory = ParallelVigraRfLazyflowClassifierFactory()
classifier = classifier_factory.create_and_train( feature_matrix,
labels,
feature_names=all_features_and_labels_df.columns[2:-1].values )
assert set(classifier.known_classes).issubset(set([1,2]))
result[0] = classifier
def test_basic(self):
# Initialize factory
factory = ParallelVigraRfLazyflowClassifierFactory(10)
# Train
classifier = factory.create_and_train(self.training_feature_matrix,
self.training_labels)
assert isinstance(classifier, ParallelVigraRfLazyflowClassifier)
assert list(classifier.known_classes) == [1, 2]
# Predict
probabilities = classifier.predict_probabilities(self.prediction_data)
assert probabilities.shape == (4, 2)
assert probabilities.dtype == numpy.float32
assert (0 <= probabilities).all() and (probabilities <= 1.0).all()
assert (numpy.argmax(probabilities, axis=-1) +
1 == self.expected_classes).all()
