hntd187
commented
9 years ago So what this seems to be to me I prototyped around with your test data in R to try and understand it. So the cycle creates basically an index for the value in each time period like
Array(1,2,3,4,5,6,7,1,2,3,4,5,6,7);
So on and so forth, then the using that it uses that for the mean to smooth.
So for a test dataset
Array(1,2,3,4,5,6,7,7,6,5,4,3,2,1); The first one would take the mean of 1 and 7 being 4, so the seasonal would then be scaled based on that mean, and the remainder re calculated as a result.
I'm hacking around with this right now, I should have a rough POC for what I mean sometime tomorrow, it'll be rough, not going to lie, but we can refine it once I have it working.
brandtg
In the R code, if periodic is true, then weighted mean smoothing is applied to the cycle subseries of the seasonal component:
This corresponds to "2.5 Post-smoothing of the seasonal" from the paper.
The
CycleSubSeriesinner class ofStlDecompositionandStlDecomposition#weightedMeanSmoothfunctions may be of use here.There's a comment in
StlDecomposition#decomposethat shows where this should go, around line 145.