使用R中的窗口进行峰选择

Question

I have a data set like this:

Just 1 table with 2 columns. The first column runs from 1 to 100 and in the second we have random numbers. for eg

x    y
1    25
2    51
3    250   
-    --
48   250
49   500
50   1000
-    ---       --and so on till
100  600

Now , I need to choose a window of first 50 rows (x = 1 to x= 50). Take the value of y corresponding to x = 50 (here y=1000 for x=50) and take the ratio of y(x = 50) to y(x = 49) ..here it is 1000 / 500 = 2... and the ratio of y(x = 50) to y(x = 48) ..here it is 1000 / 250 = 4.. and so on till y(x = 50) to y(x = 1) and then take the average of these fifty ratios.

till this part, it was easy.

after this I need to move the window by 1 position for eg from x = 2 to x= 51 and repeat everything till the window reaches x= 51 to x=100.

I was thinking of using a time series window() function or rle() function or to use the subset() function in a for loop or to use apply() function ? what is the efficient way?

Answer 1

You are looking for rollapply .

Using zoo package:

  library(zoo)
  tt <- zoo(runif(100),order.by=seq_len(100))
  rollapply(tt,width=50,
             function(x)mean(tail(x,1)/x[-length(x)]))

PS : forget the mean function.

In case you have some zeros values you can check the numerator using ifesle :

filter_x <- 
  function(x){ vv = x[-length(x)]
               vv = ifelse(vv>0,vv,1)
               mean(tail(x,1)/vv)}

rollapply(tt,width=50,filter_x)

Answer 2

Use rollapply() in the zoo package:

library(zoo)
set.seed(1)
foo <- runif(100)
foo[c(58,59)] <- 0
rollapply(foo,width=50,FUN=function(xx){
    foo <- xx[50]/xx[1:49]
    mean(foo[is.finite(foo)],na.rm=TRUE)
    }
)