%dopar% expression with the doRNG adapterEvaluates a foreach %dopar% expression with the doRNG adapter
withDoRNG(expr, substitute = TRUE, envir = parent.frame())The value of expr.
An R expression.
(logical) If TRUE, expr is substituted, otherwise not.
The environment where to evaluate expr.
Consider a function:
my_fcn <- function(n) {
y <- foreach(i = seq_len(n)) %dopar% {
stats::runif(n = 1L)
}
mean(unlist(y))
}
This function generates random numbers, but without involving doRNG, which risks generating poor randomness. If we call it as-is, with the doFuture adapter, we will get a warning about the problem:
> my_fcn(10)
[1] 0.5846141
Warning message:
UNRELIABLE VALUE: One of the foreach() iterations ('doFuture-1')
unexpectedly generated random numbers without declaring so. There is a
risk that those random numbers are not statistically sound and the overall
results might be invalid. To fix this, use '%dorng%' from the 'doRNG'
package instead of '%dopar%'. This ensures that proper, parallel-safe
random numbers are produced via the L'Ecuyer-CMRG method. To disable this
check, set option 'doFuture.rng.onMisuse' to "ignore".
>
To fix this, we use withDoRNG() as:
> withDoRNG(my_fcn(10))
[1] 0.535326
This function is useful when there is a foreach %dopar% expression that
uses the random-number generator (RNG). Such code should ideally use
%doRNG% of the doRNG package instead of %dopar%. Alternatively,
and second best, is if the code would temporarily register the doRNG
foreach adapter. If neither is done, then there is a risk that the
random numbers are not statistically sound, e.g. they might be correlated.
For what it is worth, the doFuture adapter, which is set by
registerDoFuture(), detects when doRNG is forgotten, and produced
an informative warning reminding us to use doRNG.
If you do not have control over the foreach code, you can use
withDoRNG() to evaluate the foreach statement with
doRNG::registerDoRNG() temporarily set.