Adjusting smoothing in TS Model Factory

Question

I inherited a workflow that uses a lot of TS tools.  I'm learning how they work, but one of the adjustments I would like to make is in TS Model Factory.  I believe that the smoothing is putting too much weight on the most recent time periods and would like to extend how far back the smoothing goes.

I have figured out to open the macro for TS Model Factory and believe the tool I need to adjust in the macro is the R tool, but there is a lot in it I don't understand yet.  Can someone point me to where I can adjust either the length of the lookback or the weighting?

LindonB · Accepted Answer

Hi there,

I'll share some initial ideas/info. The TS Model Factory tool basically exists to allow for multiple ETS or ARIMA models (by group) to be run at once. For ETS models, for example, an exponential smoothing model is used. Generally, the parameters for the exponential smoothing are alpha, beta and gamma. Alpha is the smoothing parameter for the level. beta is for the trend, and gamma is for the seasonality. If you feel the model is putting too much or too little emphasis on recent data, tweaking the alpha could help. A higher alpha places more emphasis on recent observations, while a lower alpha places more emphasis on historical observations. I looked at the R script behind the TS Model Factory Tool, and it does use the forecast package in R, as expected. Functions, such as predict() in this package, allow you to adjust the parameters, but normally you let the model optimize for them. (Note that there are times when you'd definitely want to adjust these parameters, but that can increase the risk of overfitting.) Thus, you'll have to adjust the script (and possibly the user interface depending on your goal) if you want to set a fixed value. Currently, setting alpha isn't an option in the forecasting tools in Alteryx.