Search code examples
rpcaeigenvalue

Find the number of eigen values that contributes relatively to a given percentage of total variance

Be λ a eigenvalue, Λ a set of eigenvalues, and p the cardinal of Λ, I'm trying to make a loop in R allowing to find n so enter image description here

Here is an example of input (L components always are decreasing) : 

> L
[1] 4.903332 4.341607 3.920309 3.716966 3.554570 3.067658 2.891222 2.553127
> sumli
[1] 28.94879
> freq <- L/sumli
> freq
[1] 0.16937952 0.14997540 0.13542221 0.12839795 0.12278821 0.10596843 0.09987368 0.08819460
Solution 
  • L <- c(4.903332, 4.341607, 3.920309, 3.716966, 3.55457, 3.067658, 
2.891222, 2.553127)

    I think we can use cumsum (cumulative sum).

    Depending on whether you want the first n such that cumulative contribution is greater than 95%, or the last n such that cumulative contribution is less than 95%, you need

    which(cumsum(L) / sum(L) > 0.95)
# [1] 8

which(cumsum(L) / sum(L) > 0.95) - 1
# or: `sum(cumsum(L) / sum(L) <= 0.95)`
# or: `max(which(cumsum(L) / sum(L) <= 0.95))`
# [1] 7

    (Your example L is not very representative as eigen values do not have a sharp decay.)