Question

I just realized that for a POSIX variable with two decimal places the following function successor() shows significant performance loss. Besides that the for loop might not be good r-style*. I was surprised that on my system POSIX with two decimals is nearly 30 times slower (20000 steps) than POSIX without decimals. POSIX with decimals is even slower than just storing the vector as character.

So is the slow performance just because of the successor() function? Or is it in general more advice able in R to store Time/Date variables as characters and just convert it when really needed?

successor <- function(z) {
y<-as.vector(z)
for(i in 1:NROW(z)) {
y[i] <- if(i == NROW(z)) NA else z[i+1]
}
return(y)
}

u<-rep(strptime("15.01.2010 10:21:52.85",format="%d.%m.%Y %H:%M:%OS"),20000) # fragments of seconds stored
v<-seq(c(ISOdate(2011,09,12)),by="min", length.out=20000) # no fragments of seconds saved
u.posix.time.small<-system.time(successor(u[1:1000]))
u.char.time.small<-system.time(successor(as.character(u[1:1000])))
u.posix.time.big<-system.time(successor(u[1:20000]))
u.char.time.big<-system.time(successor(as.character(u[1:20000])))

v.posix.time.small<-system.time(successor(v[1:1000]))
v.char.time.small<-system.time(successor(as.character(v[1:1000])))
v.posix.time.big<-system.time(successor(v[1:20000]))
v.char.time.big<-system.time(successor(as.character(v[1:20000])))
rbind(u.posix.time.small,u.posix.time.big,u.char.time.small, u.char.time.big,v.posix.time.small, v.posix.time.big, v.char.time.small,v.char.time.big)[,1:3]

*I came across the predecessor/sucessor thing when using segments(x0=x[i],x1=x[i+1], y0=y[i], y1=y[i+1]) in plot. Anyway I guess there must be another way to address the successors/predecessors since storing the values twice seems to me waste. But I am not a programmer just user.

Answer 1

In your example, u is class POSIXlt while v is class POSIXct. If you convert u to POSIXct, the timings are very similar.

POSIXlt objects are less efficient because they’re stored as a list of vectors, whereas POSIXct objects are stored as a single numeric vector.

> u <- as.POSIXct(u)
> u.posix.time.small<-system.time(successor(u[1:1000]))
> u.char.time.small<-system.time(successor(as.character(u[1:1000])))
> u.posix.time.big<-system.time(successor(u[1:20000]))
> u.char.time.big<-system.time(successor(as.character(u[1:20000])))
> rbind(u.posix.time.small,u.posix.time.big,u.char.time.small, u.char.time.big,
+ v.posix.time.small, v.posix.time.big, v.char.time.small,v.char.time.big)[,1:3]
                   user.self sys.self elapsed
u.posix.time.small      0.04     0.00    0.04
u.posix.time.big        1.91     0.01    1.92
u.char.time.small       0.01     0.00    0.02
u.char.time.big         0.29     0.02    0.32
v.posix.time.small      0.05     0.00    0.04
v.posix.time.big        1.43     0.00    1.44
v.char.time.small       0.02     0.00    0.01
v.char.time.big         0.32     0.00    0.32