On Calibration Error of Randomized Forecasting Algorithms

It has been recently shown that calibration with an error less than ∆ > 0 is almost surely guaranteed with a randomized forecasting algorithm, where forecasts are obtained by random rounding the deterministic forecasts up to ∆. We show that this error cannot be improved for a vast majority of sequences: we prove that, using a probabilistic algorithm, we can effectively generate with probability close to one a sequence “resistant” to any randomized rounding forecasting with an error much smaller than ∆. We also reformulate this result by means of a probabilistic game. Key words: Machine learning, Universal prediction, Randomized prediction, Algorithmic prediction, Calibration, Randomized rounding

• Type: Conference
• Year: 2007
• Where: ALT
• Authors: Vladimir V. V'yugin
• Keywords: Algorithmic Prediction, ALT 2007, Machine Learning, Randomized Forecasting Algorithm, Randomized Rounding Forecasting