Prior probability for classifiers
1 view (last 30 days)
In the fitcsvm method for training a SVM classifier one can set the prior for each class with the 'prior' flag. For most other classifiers this is also possible.
In the algorithm description I did not find any hint how this prior is incorporated in the training or the prediction. It is only described how the prior changes when e.g. a cost is incorporated etc.
How is the prior incorporated into training and prediction?