Predictive functions on the [0, 1] interval (pink horizontal line represents the function average on this interval). Notice that the quadratic function was limited to 2 and -2. The reason behind this decision was empirical. Pilot data showed that a smooth quadratic relation was difficult to learn. We thus decided to make it more extreme, so to make it clearer that high and low values of wind predicted rain, while values that were close to 50% were predictive of no-rain.