elephant
elephant copied to clipboard
make variables remember their history
elephant 
Elephants never forget!
elephant is a quick experiment on how to make variables remember their history.
Use := to record a call.
Installation
Install with:
remotes::install_github("moodymudskipper/elephant")
Example
library(elephant)
We build an elephant object by using :=
x := 1
y := 3
z := x + y
x := x * 2
bar <- x + y + z
foo := x + sqrt(z) + bar
x
#> [1] 2
#> x <- x * 2 # numeric, value: 2
#> x <- 1 # numeric, value: 1
y
#> [1] 3
#> y <- 3 # numeric, value: 3
z
#> [1] 4
#> z <- x + y # numeric, value: 4
#> x <- 1 # numeric, value: 1
#> y <- 3 # numeric, value: 3
bar
#> [1] 9
foo
#> [1] 13
#> foo <- x + sqrt(z) + bar # numeric, value: 13
#> x <- x * 2 # numeric, value: 2
#> x <- 1 # numeric, value: 1
#> z <- x + y # numeric, value: 4
#> x <- 1 # numeric, value: 1
#> y <- 3 # numeric, value: 3
#> bar # numeric, value: 9
Here we see bar is not an elephant object because it doesn’t use :=
and +.elephant doesn’t preserve class and attributes.
However baz <- x would just copy the object, so in this case it’s
advised to do either baz := x or baz <- forget(x)
# not good
baz <- x
baz
#> [1] 2
#> x <- x * 2 # numeric, value: 2
#> x <- 1 # numeric, value: 1
# good
baz := x
baz
#> [1] 2
#> baz <- x # numeric, value: 2
#> x <- x * 2 # numeric, value: 2
#> x <- 1 # numeric, value: 1
# good
baz <- forget(x)
baz
#> [1] 2
We can access the variables used to compute our elephant object by using
list_calves() or calf().
list_calves(foo)
#> $x
#> [1] 2
#> x <- x * 2 # numeric, value: 2
#> x <- 1 # numeric, value: 1
#>
#> $z
#> [1] 4
#> z <- x + y # numeric, value: 4
#> x <- 1 # numeric, value: 1
#> y <- 3 # numeric, value: 3
#>
#> $bar
#> [1] 9
list_calves(foo, "z")
#> $x
#> [1] 1
#> x <- 1 # numeric, value: 1
#>
#> $y
#> [1] 3
#> y <- 3 # numeric, value: 3
calf(foo, "z")
#> [1] 4
#> z <- x + y # numeric, value: 4
#> x <- 1 # numeric, value: 1
#> y <- 3 # numeric, value: 3
calf(foo, "z", "x")
#> [1] 1
#> x <- 1 # numeric, value: 1
Debugging with elephant
If an elephant calls fails, its memory is preserved and can be recovered
using poach(), see the example below :
test <- function(){
x := 1
y := 3
z := x * 2
bar <- x + y + z
foo := x + sqrt(z) + bar * letters # <- problematic call!
}
test()
#> Elephant was killed in foo := x + sqrt(z) + bar * letters
#> Use poach() to collect its memory
#> Error in bar * letters: non-numeric argument to binary operator
poach()
#> [1] NA
#> foo <- x + sqrt(z) + bar * letters # logical, value: NA
#> x <- 1 # numeric, value: 1
#> z <- x * 2 # numeric, value: 2
#> x <- 1 # numeric, value: 1
#> bar # numeric, value: 6
#> letters # character, length: 26
# extract the bar variable as used by foo, for further investigation
foo_poached <- poach()
calf(foo_poached, "letters")
#> [1] "a" "b" "c" "d" "e" "f" "g" "h" "i" "j" "k" "l" "m" "n" "o" "p" "q" "r" "s"
#> [20] "t" "u" "v" "w" "x" "y" "z"
# we can go deeper
calf(foo_poached, "z", "x")
#> [1] 1
#> x <- 1 # numeric, value: 1
Benchmark
:= has a small overhead, it shouldn’t slow your code much most of the
time:
bench::mark(
elephant_assignment = (foo := x + sqrt(z) + bar),
standard_assignment = (foo <- x + sqrt(z) + bar),
check = FALSE)
#> # A tibble: 2 x 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 elephant_assignment 59us 65.9us 14454. 0B 18.6
#> 2 standard_assignment 27.9us 31.6us 29790. 0B 17.9
However it wouldn’t not be well advised, nor really that useful, to use it in a loop growing an object as below.
x := 1
for (i in 1:4) {
x:= x + 1
}
x
#> [1] 5
#> x <- x + 1 # numeric, value: 5
#> x <- x + 1 # numeric, value: 4
#> x <- x + 1 # numeric, value: 3
#> x <- x + 1 # numeric, value: 2
#> x <- 1 # numeric, value: 1
It’s especially true for bigger objects as we’d be keeping in memory all previous values that would have been flushed by the garbage collector if we hadn’t used elephant.