My personal R package

Resource:	Travis CI
Platform:	Linux
R CMD check

Because I want to share pieces of code that are helpful to me.

I like this filter quite a lot and used it in my recent paper to smooth eigenvectors from several genomic matrices. It's simple (k iterations of a simple moving average), fast and the degree of smoothness can be tuned easily. The two main parameters are:

• k: number of iterations (k). Should be based on prior information if possible
• w: smoothing window width (odd integer number).
• optimal: should the optimality criterion discussed below be used? If set to TRUE, then k does not need to be specified since it will be chosen optimally.

In order to pick a reasonable number of iterations, I implemented an intuitive optimality criterion. Basically, I want the iterative process to stop at k iterations when the difference between the (k+1)-th and the k-th smoothed vectors is less than a specified threshold. I quantify difference as the sums of squares of the difference vector, i.e. SS(x_(k+1)-x_k), divided by the sum of squares of the k-th vector, that is SS(x_k). If the difference is less than the parameter tolerance, set by default to 0.05, then the iterative process stops. Below is an example of a vector before and after smoothing with an optimal number of iterations k=2.

library(jipito)
mybarplot(x)
mybarplot(kz(x, w=3, optimal=TRUE))