🎉 🎉 RAPIDMINER 9.10 IS OUT!!! 🎉🎉

Download the latest version helping analytics teams accelerate time-to-value for streaming and IIOT use cases.


"Rate decaying a (time series) variable in rapid miner?"

villepohjanheimvillepohjanheim Member Posts: 2 Contributor I
edited May 2019 in Help
I've been trying to get my head around how to do a sort of rate decay algorithm in RapidMiner (5). Can anyone give me pointers on what way should I think solving this problem in RapidMiner?

The setup: As a source I have spending data by date and I'm trying to transform the raw figures into figures that more correctly describe the effect of the spending (think advertising spending). As an example let's think of spending that has a decay rate of 1/2. In this specific way of thinking this means that if I invest 100, it will have an effect of 50 now and 25 tomorrow and so on. More generally: x(t) - x(t-1) where x(t) = x(0) * (1-decay^t)

Each point in time has its own spending so simply running the above formula on a single figure is not enough. So as raw figures t=1 spending=100; t=2 spending=0; t=3 spending=50, t=4 spending=100 etc)

(If you'd rather think of the whole problem, I am trying to find a rate decay that will create a variable that best correlates with my output/profit variable. First step in this analysis is decaying the spending variable...)


  • landland RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 2,531   Unicorn
    I think your problem will be solved after you window your time series. This is the standard approach in RapidMiner to tackle such problems:
    If you have an (univariate, for simplicity) time series and you window it in a way that you note for each timepoint the values for the last 5 time points, then a time series like "1 2 3 4 5 6 7 8 9" is changed to a set of examples like this:
    1 2 3 4 [5]
    2 3 4 5 [6]
    3 4 5 6 [7]
    4 5 6 7 [8]
    5 6 7 8 [9]
    If you apply a regression technique on this, the decay rates for each point in time will be automatically adapted.

    With kind regards,
      Sebastian Land
Sign In or Register to comment.