Read on Twitter
Our recent primer paper for using PCA and functional PCA (fPCA) in biomechanics is now out! And here's a quick thread, with some links to software/demo code (for application using R and Matlab)! https://www.sciencedirect.com/science/article/pii/S0021929020305303 1/n
This was fun to write for a few reasons
Firstly, we were able to show that PCA applied to time-points is a rudimentary form of fPCA, and can be rationalised within the FDA framework of statistical processes. A simple message, but nice to clarify and demonstrate 2/n
Firstly, we were able to show that PCA applied to time-points is a rudimentary form of fPCA, and can be rationalised within the FDA framework of statistical processes. A simple message, but nice to clarify and demonstrate 2/n
fPCA does however open up more possibilities, particularly for control over smoothing, with us demonstrating smoothing of fPCs using motion sickness data ( https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0144466). Other benefits include registration + exploring derivatives... but weren't covered in detail 3/n
Secondly, we created some tutorials for applying PCA and fPCA in both Matlab and R, both of which are available here: https://github.com/johnwarmenhoven/PCA-FPCA (will try to have these up on the website with the article as well as Github) 4/n
Thirdly, because @gunning_edward is like a statistics/data-viz John Frusciante... plugging FDA + ggplot into some sort of effects pedal to create this gif (reconstructing original curves using the fPCs). If you like great visualisations in biomechanics give him a follow! 5/n
And finally because of writing it as a multi-disciplinary group. It was really great that @NormaBargary, @gileshooker, @domliebl and @gunning_edward were interested in a paper that was simple (in terms of what they are used to)... but practically useful for biomechanics. 6/n
And very handy having Drew and @mrobbo18's expertise, for finding the balance of where to put information (and what to include/exclude) for a short methods piece. I learned more from writing this article, than any other because of how varied these perspectives were. 7/n
And one last comment... 35 years ago in 1985, the Dutch biomechanist Herman Woltring wrote a methods paper in Human Movement Science on spline based smoothing for biomechanics data... https://www.sciencedirect.com/science/article/abs/pii/0167945785900041 8/n
In that paper he mentioned that B-splines could be better than the quintic splines he trialled, and recommended exploring them in the future. So... hopefully the tutorials applying B-splines with fPCA with this article are a small step towards them being used more easily! 9/9
