CODEWITHSUNDEEP
rzoorollapply
Andrew Bannerman
Andrew Bannerman

Reputation: 1305

Rolling lagged differences

Ok so I am looking to create rolling lagged differences in R. 

vec <- c(43.79979, 44.04865, 44.17308, 44.54638, 44.79524, 44.79524, 44.79524, 44.42195, 44.54638, 44.79524, 44.42195, 43.30206, 43.30206, 43.17764, 43.30206)

> length(vec)
[1] 15

This is what I have tried so far: 

vec1 <- rollapply(vec, width = 2,  fill = NA, FUN = diff)

This gives this output: 

[1]  0.24886  0.12443  0.37330  0.24886  0.00000  0.00000 -0.37329  0.12443  0.24886 -0.37329 -1.11989  0.00000 -0.12442  0.12442       NA

> length(vec1)
[1] 15

Note we have an NA value in element 15.

So I want to do this diff in lags for say lags 1,2 and 3... So the above code doesn't cater for this, so I try the below: 

lag1 <- diff(vec, lag = 1, differences = 1, arithmetic = TRUE, na.pad = TRUE)
lag2 <- diff(vec, lag = 2, differences = 1, arithmetic = TRUE, na.pad = TRUE)
lag3 <- diff(vec, lag = 3, differences = 1, arithmetic = TRUE, na.pad = TRUE)
length(lag1)
length(lag2)
length(lag3)

The result of this: 

> lag1
 [1]  0.24886  0.12443  0.37330  0.24886  0.00000  0.00000 -0.37329  0.12443  0.24886 -0.37329 -1.11989  0.00000 -0.12442  0.12442
> lag2
 [1]  0.37329  0.49773  0.62216  0.24886  0.00000 -0.37329 -0.24886  0.37329 -0.12443 -1.49318 -1.11989 -0.12442  0.00000
> lag3
 [1]  0.74659  0.74659  0.62216  0.24886 -0.37329 -0.24886  0.00000  0.00000 -1.24432 -1.49318 -1.24431  0.00000
> length(lag1)
[1] 14
> length(lag2)
[1] 13
> length(lag3)
[1] 12

Notice that when do the lagged difference above... it places the diff result on the line that it subtracted the value on... so it took our current value - lagged value. It places the diff result on the lagged value position. We then lose the length of the vector. I want to actually place the diff - lagged result on the start number (diff) and place leading NA's to account for the missing values at the start of the data set.

Using lag 2 as en example, this is my desired result: 

> lag2
 [1]  NA    NA 0.37329  0.49773  0.62216  0.24886  0.00000 -0.37329 -0.24886  0.37329 -0.12443 -1.49318 -1.11989 -0.12442  0.00000

Does anyone know a way on how to correct this??

To maybe explain a little more:

this is is the start of the vector: 

vec <- c(43.79979, 44.04865, 44.17308.....

So if we do a lagged 2 difference...

We take the 3rd element... 44.17308 - 43.79979 = the result of 0.37329.

So I want to have NA NA 0.37329

Instead of placing 0.37329 on the first position in the new lag2 vector.

Upvotes: 3

Views: 1525

Answers (2)

elcortegano
elcortegano

Reputation: 2684

For those looking a tidyverse solution, one option is to use dplyr::lag, which I find more intuitive than that with base::apply.

    vec - dplyr::lag(vec, n = 2)

So the idea is basically to generate a second vector with the positions n-lagged, and just substracting the two vectors without further complication, making the most of vectorized functions in R.

Upvotes: 1

catastrophic-failure
catastrophic-failure

Reputation: 3905

Just like in Zoo lag diff back in data frame

vec = c(43.79979, 44.04865, 44.17308, 44.54638, 44.79524, 44.79524, 44.79524, 44.42195, 44.54638, 44.79524, 44.42195, 43.30206, 43.30206, 43.17764, 43.30206)
require(zoo)

apply(lag(zoo(vec), c(-2,0), na.pad = TRUE), 1L, diff)

#> apply(lag(zoo(vec), c(-2,0), na.pad = TRUE), 1L, diff)
#       1        2        3        4        5        6        7        8        9       10       11       12       13       14       15 
#      NA       NA  0.37329  0.49773  0.62216  0.24886  0.00000 -0.37329 -0.24886  0.37329 -0.12443 -1.49318 -1.11989 -0.12442  0.00000

On May 10th 2018 it was pointed to me by @thistleknot (thanks!) that dplyr masks stats's own lag generic. Therefore make sure you don't have dplyr attached, or instead run stats::lag explicitly, otherwise my code won't run.

I think I found the culprit: github.com/tidyverse/dplyr/issues/1586 answer: This is a natural consequence of having lots of R packages. Just be explicit and use stats::lag or dplyr::lag

Upvotes: 3

Related Questions