I would like to implement a "fast attack / slow decay" (peak detect with exponential decay) filter on a time series ts (a column in a pandas dataframe), described