How to get the maximum value of N past observations

Question

Suppose y_t is a time series. What I want is x_t = max(y_t,y_{t -1},...,y_{t - N}). My code is

x<-unlist(lapply(seq_along(y),function(i)max(y[seq.int(max(1L,i-N),i)])))

This code uses lapply and there are repetitive reads, so I think it isn't taking full advantage of SIMD features of the processor. Is there a faster solution?

Answer 1

We can use roll_max from RcppRoll

library(RcppRoll)
roll_max(v1, n=3)

data

v1 <- 1:10