Description
I know it doesnt make sense as you would ideally want a separate downsample for each column in your dataframe but what about some strategy like:
pick your max number of points to retain per column -> run lttb for each timestamp, col and retain the indexes in another dataframe so that would be a NxM matrix with no nulls -> k mediods on all columns -> find mean delta of the indexes in order ascending -> select indexes in original dataframe every delta and resample at the delta?
Could replace the lttb part with anything really but also wondering what an ideal pipeline would look like? The goal is to take a dataframe with one timestamp index value and many columns and produce a dataframe with the same number of columns and ideally less rows. Definitely dont gain all the benefits as you would by doing this individually for each column but a large benefit is gained by retaining a rectangular matrix input and output