Skip to main content

Drop Uncertain Predictions

Synopsis​

This operator sets all predictions to 'unknown' (missing value) if the corresponding confidence is less than the specified minimum confidence. This operator is used for dropping predictions with low confidence values.

Description​

The Drop Uncertain Predictions operator expects a labeled ExampleSet i.e. an ExampleSet with label and prediction attributes along with prediction confidences. The minimum confidence threshold is specified through the min confidence parameter. All those predictions of the given ExampleSet are dropped where the corresponding prediction confidence is below the specified threshold. Suppose an ExampleSet with two possible classes 'positive' and 'negative'. If the min confidence parameter is set to 0.700, all the examples that were predicted as 'positive' but their corresponding 'confidence (positive)' value is less than 0.700 are classified as missing values. Similarly the label value is set to missing value for all those examples that were predicted as 'negative' but their corresponding confidence '(negative)' value is less than 0.700. This operator also allows you to define different minimum confidence thresholds for different classes through the min confidences parameter.

Input​

example set input​

This input port expects a labeled ExampleSet. It is the output of the Apply Model operator in the attached Example Process. The output of other operators can also be used as input if it is a labeled ExampleSet.

Output​

example set output​

The uncertain predictions are dropped and the resultant ExampleSet is delivered through this port.

original​

The ExampleSet that was given as input is passed without changing to the output through this port. This is usually used to reuse the same ExampleSet in further operators or to view the ExampleSet in the Results Workspace.

Parameters​

Class handling​

This parameter specifies the mode of class handling which defines if all classes are handled equally or if individual class thresholds are set.

  • balanced: In this case all classes are handled equally i.e. the same confidence threshold is applied on all possible values of the label. The minimum confidence threshold is specified through the min confidence parameter.
  • unbalanced: In this case classes are not handled equally i.e. different confidence thresholds can be specified for different classes through the min confidences parameter.

Min confidence​

This parameter is only available when the class handling parameter is set to 'balanced'. This parameter sets the minimum confidence threshold for all the classes. Predictions below this confidence will be dropped.

Min confidences​

This parameter is only available when the class handling parameter is set to 'unbalanced'. This parameter specifies individual thresholds for classes. Predictions below these confidences will be dropped.