class PredictionsMonitoring(object):
def __init__(self):
self.predictions = None
self.proba_barplot = None
def add_fold(self, predictions):
if self.predictions is None:
ids = Ids(deepcopy(predictions.ids.ids),
deepcopy(predictions.ids.idents),
deepcopy(predictions.ids.timestamps))
self.predictions = Predictions(deepcopy(predictions.values), ids,
predictions.info.multiclass,
deepcopy(predictions.all_probas),
deepcopy(predictions.probas),
deepcopy(predictions.scores),
deepcopy(predictions.ground_truth))
else:
self.predictions.union(predictions)
def final_computations(self):
# PredictionsBarplots only for probabilist binary models
pred_info = self.predictions.info
if not pred_info.multiclass and pred_info.with_probas:
self.proba_barplot = ProbaBarplot(pred_info.with_ground_truth)
self.proba_barplot.add_fold(self.predictions)
def display(self, directory):
if self.proba_barplot is not None:
self.proba_barplot.display(directory)
def apply_pipeline(self, instances):
num_instances = instances.num_instances()
if num_instances == 0:
return Predictions([], [], [], [], instances.ids)
features = instances.features.get_values()
predictions = self._predict(features)
all_probas, probas = self._get_predicted_probas(
features, num_instances)
scores = self._get_predicted_scores(features, num_instances)
return Predictions(list(predictions), all_probas, probas, scores,
instances.ids)
def add_fold(self, predictions):
if self.predictions is None:
ids = Ids(deepcopy(predictions.ids.ids),
deepcopy(predictions.ids.idents),
deepcopy(predictions.ids.timestamps))
self.predictions = Predictions(deepcopy(predictions.values), ids,
predictions.info.multiclass,
deepcopy(predictions.all_probas),
deepcopy(predictions.probas),
deepcopy(predictions.scores),
deepcopy(predictions.ground_truth))
else:
self.predictions.union(predictions)
def apply_pipeline(self, instances):
num_instances = instances.num_instances()
if num_instances == 0:
return Predictions([], [], [], [], instances.ids)
features = instances.features.get_values()
preprocessed_features = self.scaler.transform(features)
predicted_scores = np.apply_along_axis(predict_score, 1,
preprocessed_features, self.c,
self.r)
predicted_labels = predicted_scores > 0
predicted_proba = [None for i in range(num_instances)]
all_predicted_proba = [None for i in range(num_instances)]
return Predictions(predicted_labels, all_predicted_proba,
predicted_proba, predicted_scores, instances.ids)
def apply_pipeline(self, instances):
num_instances = instances.num_instances()
if num_instances == 0:
return Predictions([], instances.ids, False)
features = instances.features.get_values()
preprocessed_features = self.scaler.transform(features)
predicted_scores = np.apply_along_axis(predict_score, 1,
preprocessed_features, self.c,
self.r)
predicted_labels = predicted_scores > 0
return Predictions(predicted_labels,
instances.ids,
False,
scores=predicted_scores)
def apply_pipeline(self, instances):
num_instances = instances.num_instances()
if num_instances == 0:
return Predictions([], instances.ids, self.conf.multiclass)
features = instances.features.get_values()
predictions = self._predict(features)
all_probas, probas = self._get_predicted_probas(
features, num_instances)
scores = self._get_predicted_scores(features, num_instances)
return Predictions(predictions,
instances.ids,
self.conf.multiclass,
all_probas=all_probas,
probas=probas,
scores=scores)
def _predict_matrix(self, matrix, instances_ids):
values = self._predict_values(matrix)
all_probas, probas = self._predict_probas(matrix)
all_scores, scores = self._predict_scores(matrix)
return Predictions(values, instances_ids, self.conf.multiclass,
all_probas=all_probas, probas=probas,
all_scores=all_scores, scores=scores)
def _detection(self, kind, classifier, instances, fold_id):
global_exp, detection_exps = self._create_detection_exp(
kind, classifier.conf, fold_id=fold_id)
if not all(detection_exp.exp_conf.dataset_conf.has_ground_truth ==
detection_exps[0].exp_conf.dataset_conf.has_ground_truth
for detection_exp in detection_exps):
raise InvalidValidationDatasets(
'All the test datasets must contain ground truth '
'annotations, or none of them.')
if fold_id is None:
exp = global_exp if global_exp is not None else detection_exps[0]
self._set_detection_exp(kind, exp)
predictions = None
prediction_time = None
for detection_exp in detection_exps:
detection_exp.run(instances, classifier)
if predictions is None:
predictions = Predictions.deepcopy(detection_exp.predictions)
predictions = detection_exp.predictions
prediction_time = detection_exp.prediction_time
else:
predictions.union(detection_exp.predictions)
prediction_time += detection_exp.prediction_time
if global_exp is not None:
global_exp.set_predictions(predictions, prediction_time)
return predictions, prediction_time
def add_fold(self, predictions):
self.num_added_folds += 1
if self.num_added_folds > self.num_folds:
raise AddTooManyFolds()
if self.predictions is None:
if self.num_folds > 1:
self.predictions = Predictions.deepcopy(predictions)
else:
self.predictions = predictions
else:
self.predictions.union(predictions)
def _predict_matrix(self, matrix, instances_ids):
all_probas, probas = self._predict_probas(matrix)
if all_probas is None and probas is None:
all_scores, scores = self._predict_scores(matrix)
if all_scores is None and scores is None:
values = self._predict_values(matrix)
else:
values = self._predict_from_scores(matrix, all_scores, scores)
else:
all_scores, scores = None, None
values = self._predict_from_probas(matrix, all_probas, probas)
if probas is not None:
probas = probas[:, 1]
return Predictions(values,
instances_ids,
self.conf.multiclass,
all_probas=all_probas,
probas=probas,
all_scores=all_scores,
scores=scores)
def add_fold(self, predictions):
if self.predictions is None:
self.predictions = Predictions.deepcopy(predictions)
else:
self.predictions.union(predictions)
def apply_pipeline(self, instances):
num_instances = instances.num_instances()
if num_instances == 0:
return Predictions(np.array([]), instances.ids,
self.conf.multiclass)
return self._predict(instances.features, instances.ids)