Calibration, Predictiveness, Randomness and Remorse

View a PDF of the paper titled 4 Sides of Forecast Felicity: Calibration, Predictiveness, Randomness and Remorse, by Rabanus Derr and Robert C. Williamson

View PDF
HTML (experimental)

Summary:Machine studying is about forecasting. Forecasts, nevertheless, acquire their usefulness solely by their analysis. Machine studying has historically targeted on kinds of losses and their corresponding remorse. Presently, the machine studying neighborhood regained curiosity in calibration. On this work, we present the conceptual equivalence of calibration and remorse in evaluating forecasts. We body the analysis downside as a recreation between a forecaster, a gambler and nature. Placing intuitive restrictions on gambler and forecaster, calibration and remorse naturally fall out of the framework. As well as, this recreation hyperlinks analysis of forecasts to randomness of outcomes. Random outcomes with respect to forecasts are equal to good forecasts with respect to outcomes. We name these twin points, calibration and remorse, predictiveness and randomness, the 4 aspects of forecast felicity.