(AI Studio Core)
Synopsis
This operator creates a user defined threshold for crisp classification based on the prediction confidences (soft predictions). This threshold can be applied by using the Apply Threshold operator.Description
The threshold parameter specifies the required threshold. The first class and second class parameters are used for specifying the classes of the ExampleSet that should be considered as first and second class respectively. The threshold created by this operator can be applied on the labeled ExampleSet using the Apply Threshold operator. Should it occur that the confidence for the second class is greater than the given threshold then the prediction is set to this second class otherwise it is set to the first class. This can be easily understood by studying the attached Example Process.
The Apply Threshold operator applies the given threshold to a labeled ExampleSet and maps a soft prediction to crisp values. The threshold is provided through the threshold port. Mostly the Create Threshold operator is used for creating thresholds before they are applied using the Apply Threshold operator.
Among various classification methods, there are two main groups of methods: soft and hard classification. In particular, a soft classification rule generally estimates the class conditional probabilities explicitly and then makes the class prediction based on the largest estimated probability. In contrast, hard classification bypasses the requirement of class probability estimation and directly estimates the classification boundary.
Output
output (Threshold Model)This port delivers the threshold. This threshold can be applied on a labeled ExampleSet by using the Apply Threshold operator.
Parameters
- thresholdThis parameter specifies the threshold of the prediction confidence. It should be in range 0.0 to 1.0. If the prediction confidence for the second class is greater than this threshold the prediction is set to second class (i.e. the class specified through the second class parameter) otherwise it is set to the first class(i.e. the class specified through the first class parameter).
- first classThis parameter specifies the class which should be considered as the first class.
- second classThis parameter specifies the class which should be considered as the second class.
Tutorial Processes
Creating and Applying thresholds
This Example Process starts with a Subprocess operator. This subprocess provides the labeled ExampleSet. Double-click on the Subprocess operator to see what is happening inside the subprocess although it is not directly relevant to the use of the Create Threshold operator. In the subprocess, the k-NN classification model is learned and applied on different samples of the 'Weighting' data set. The resultant labeled ExampleSet is output of this subprocess. A breakpoint is inserted after this subprocess so that you can have a look at the labeled ExampleSet before the application of the Create Threshold and Apply Threshold operators. You can see that the ExampleSet has 20 examples. 11 of them are predicted as 'positive' and the remaining 9 examples are predicted as 'negative'. If you sort the results according to the confidence of positive prediction, you will easily see that among 11 examples predicted as 'positive', 3 examples have confidence 0.600, 4 examples have confidence 0.700, 3 examples have confidence 0.800 and 1 example has confidence 0.900.
Now let us have a look at what is happening outside the subprocess. The Create Threshold operator is used for creating a threshold. The threshold parameter is set to 0.700 and the first class and second class parameters are set to 'negative' and 'positive' respectively. A breakpoint is inserted here so that you can see the threshold in the Results Workspace. This statement in the Results Workspace explains everything:
if confidence(positive) > 0.7 then positive; else negative
This statement means that if confidence(positive) is greater than 0.7 then the class should be predicted as positive otherwise it should be predicted as negative. In a general form this statement would look something like this:
if confidence(second) > T then second; else first.
where T, second and first are the values of the threshold, second class and first class parameters respectively.
This threshold is applied on the labeled ExampleSet using the Apply Threshold operator. We know that when the Apply Threshold operator is applied on an ExampleSet there are two possibilities: if the confidence for the second class is greater than the given threshold the prediction is set to second otherwise to the first class. In this process, if the confidence for the second class i.e. 'positive' (class specified in the second class parameter of the Create Threshold operator) is greater than the given threshold i.e. 0.700 (threshold specified in the threshold parameter of the Create Threshold operator) the prediction is set to 'positive' otherwise it is set to 'negative'. In the labeled ExampleSet only 4 examples had confidence (positive) greater than 0.700. When the Apply Threshold operator is applied only these 4 examples are assigned 'positive' predictions and all other examples are assigned 'negative' predictions.
