Improvements to allow sending a batch of predictions at once, using the exponential smoothing predictor