look at Greenwald-Khanna (and the followon work). It adapts the bucket size to minimize the total error (and the number of buckets is proportional to log-epsilo

look at Greenwald-Khanna (and the followon work). It adapts the bucket size to minimize the total error (and the number of buckets is proportional to log-epsilon I think). with all the competition and 'innovation', you would think the data dogs of this world would grow up beyond time series.