def predict_operating(self, input_data, operating_point=None):
"""Computes the prediction based on a user-given operating point.
"""
kind, threshold, positive_class = parse_operating_point( \
operating_point, ["probability"], self.class_names)
predictions = self.predict_probability(input_data, False)
position = self.class_names.index(positive_class)
if predictions[position][kind] > threshold:
prediction = predictions[position]
else:
# if the threshold is not met, the alternative class with
# highest probability or confidence is returned
predictions.sort( \
key=cmp_to_key( \
lambda a, b: self._sort_predictions(a, b, kind)))
prediction = predictions[0 : 2]
if prediction[0]["category"] == positive_class:
prediction = prediction[1]
else:
prediction = prediction[0]
prediction["prediction"] = prediction["category"]
del prediction["category"]
return prediction
def predict_operating(self, input_data,
missing_strategy=LAST_PREDICTION,
operating_point=None):
"""Computes the prediction based on a user-given operating point.
"""
# only probability is allowed as operating kind
operating_point.update({"kind": "probability"})
kind, threshold, positive_class = parse_operating_point( \
operating_point, OPERATING_POINT_KINDS, self.class_names)
predictions = self.predict_probability(input_data,
missing_strategy, False)
position = self.class_names.index(positive_class)
if predictions[position][kind] > threshold:
prediction = predictions[position]
else:
# if the threshold is not met, the alternative class with
# highest probability or confidence is returned
predictions.sort( \
key=cmp_to_key( \
lambda a, b: self._sort_predictions(a, b, kind)))
prediction = predictions[0: 2]
if prediction[0]["category"] == positive_class:
prediction = prediction[1]
else:
prediction = prediction[0]
prediction["prediction"] = prediction["category"]
del prediction["category"]
return prediction
def predict_operating(self,
input_data,
by_name=True,
operating_point=None,
compact=False):
"""Computes the prediction based on a user-given operating point.
"""
kind, threshold, positive_class = parse_operating_point( \
operating_point, ["probability"], self.class_names)
predictions = self.predict_probability(input_data, by_name, compact)
position = self.class_names.index(positive_class)
if predictions[position][kind] < threshold:
# if the threshold is not met, the alternative class with
# highest probability or confidence is returned
prediction = sorted(predictions, key=lambda x: -x[kind])[0:2]
if prediction[0]["category"] == positive_class:
prediction = prediction[1]
else:
prediction = prediction[0]
else:
prediction = predictions[position]
prediction["prediction"] = prediction["category"]
del prediction["category"]
return prediction
def predict_operating(self,
input_data,
missing_strategy=LAST_PREDICTION,
operating_point=None):
"""Computes the prediction based on a user-given operating point.
"""
kind, threshold, positive_class = parse_operating_point( \
operating_point, OPERATING_POINT_KINDS, self.class_names)
try:
predict_method = None
predict_method = getattr(self, "predict_%s" % kind)
predictions = predict_method(input_data, missing_strategy, False)
position = self.class_names.index(positive_class)
except KeyError:
raise ValueError("The operating point needs to contain a valid"
" positive class, kind and a threshold.")
if self.regression:
prediction = predictions
else:
position = self.class_names.index(positive_class)
if predictions[position][kind] > threshold:
prediction = predictions[position]
else:
# if the threshold is not met, the alternative class with
# highest probability or confidence is returned
predictions.sort( \
key=cmp_to_key( \
lambda a, b: self._sort_predictions(a, b, kind)))
prediction = predictions[0:2]
if prediction[0]["category"] == positive_class:
prediction = prediction[1]
else:
prediction = prediction[0]
prediction["prediction"] = prediction["category"]
del prediction["category"]
return prediction
