calibration-example

This repository hosts a small example for calibrating binary classifiers. In particular, it shows how to use Matlab's implementation of Platt scaling to calibrate the outputs of an SVM.

Running the example

Download (and unzip) the Matlab version of the USPS data set. You can use the provided script:

    ./getUSPS.sh

In matlab, run the toplevel script:

    usps_example

References

• Platt "Probabilistic outputs for support vector machines and comparison to regularized likelihood methods," 1999.
• Zadrozny and Elkan "Transforming classifier scores into accurate multiclass probability estimates," 2002.
• Niculescu-Mizil and Caruana "Predicting good probabilities with supervised learning," 2005.
• Tight subplot Matlab script