« Latest Redmonk and Tiobe language rankings for R | Main | Explore New Zealand's Tourist Industry with R and Shiny »

February 23, 2016


Feed You can follow this conversation by subscribing to the comment feed for this post.

How does this decomposition work?

Thank You.

I know you're trying to use base R, but Twitter's anomaly detection package I think would do wonders for this dataset!

thank you for intresting example. I have a feeling, that Data are having also multiple seasonal components (sommer/winter seasonality). Would it be intresting to try Fourier or Wavelet transformations?
It looks like that Xbox and TV usage having some Weekly and Monthly correlation.


Thanks for the comments.

Royi - the decomposition uses the LOESS algorithm built in to R. It's a non-parametric least squares regression. It basically fits a low-degree polynomial to each data point using a weighted least squares method. The Wikipedia article is a good starting point with references for more detail - https://en.wikipedia.org/wiki/Local_regression

Amit - I agree. The anomaly detection package helps a lot!

Igor - thanks for your comments. I have used both a Fourier and Wavelet transformations in looking at these data sets to help reveal multiple seasonality. However, I find that simple decompositions lend themselves to easier explanations (useful when reporting results to non-data scientists/statisticians), even though they may not have the fidelity that Fourier or Wavelet transformations have...

The comments to this entry are closed.

Search Revolutions Blog

Got comments or suggestions for the blog editor?
Email David Smith.
Follow revodavid on Twitter Follow David on Twitter: @revodavid