Does argument `parallelseed` in `futuremice()` work as expected?

[stefvanbuuren]

I am unsure whether I understand the intended behaviour of parallelseed. I would think that specifying the same parallelseed at two identical calls would yield the same imputations, as happens with seed in mice().

The test script below for mice 3.14.9 shows that imputes are different. Is my understanding correct? Any ideas how to fix?

library(mice, warn.conflicts = FALSE)
library(testthat)

context("mice: seed")

imp1 <- mice(nhanes2, m = 2, seed = 123, print = FALSE)
imp2 <- mice(nhanes2, m = 2, seed = 123, print = FALSE)
test_that("same seed yields same imputations", {
  expect_identical(imp1[-11], imp2[-11])
})
#> Test passed 🎉

context("futuremice: parallelseed")

imp1 <- futuremice(nhanes2, m = 4, n.core = 2, parallelseed = 1)
imp2 <- futuremice(nhanes2, m = 4, n.core = 2, parallelseed = 1)
test_that("same parallelseed yields same imputations", {
  expect_identical(imp1[-11], imp2[-11])
})
#> ── Failure ('<text>:17'): same parallelseed yields same imputations ────────────
#> imp1[-11] not identical to imp2[-11].
#> Component "imp": Component "bmi": Component "1": Mean relative difference: 0.1554468
#> Component "imp": Component "bmi": Component "2": Mean relative difference: 0.1348936
#> Component "imp": Component "bmi": Component "3": Mean relative difference: 0.1516
#> Component "imp": Component "bmi": Component "4": Mean relative difference: 0.1243956
#> Component "imp": Component "hyp": Component "1": 2 string mismatches
#> Component "imp": Component "hyp": Component "2": 4 string mismatches
#> Component "imp": Component "hyp": Component "3": 2 string mismatches
#> Component "imp": Component "hyp": Component "4": 4 string mismatches
#> Component "imp": Component "chl": Component "1": Mean relative difference: 0.2008149
#> ...
#> Error in `reporter$stop_if_needed()`:
#> ! Test failed

#> Backtrace:
#>      ▆
#>   1. └─testthat::test_that(...)
#>   2.   └─withr (local) `<fn>`(`<env>`)
#>   3.     ├─base::tryCatch(...)
#>   4.     │ └─base (local) tryCatchList(expr, classes, parentenv, handlers)
#>   5.     │   └─base (local) tryCatchOne(expr, names, parentenv, handlers[[1L]])
#>   6.     │     └─base (local) doTryCatch(return(expr), name, parentenv, handler)
#>   7.     └─base::eval(handler$expr, handler$envir)
#>   8.       └─base::eval(handler$expr, handler$envir)
#>   9.         └─reporter$stop_if_needed()
#>  10.           └─rlang::abort("Test failed")

^{Created on 2022-11-09 with reprex v2.0.2}

[gerkovink]

Nope. Does not seem to be platform dependent. Check fail on:

• x86_64-pc-linux-gnu (64-bit)
• aarch64-apple-darwin20 (64-bit)
• x86_64-apple-darwin17.0 (64-bit).

[gerkovink]

Will pick this up

[gerkovink]

Argument parallelseed experienced override by the internal mice call to set.seed(NULL) introduced in v3.14.4. Fixed in #515

[gerkovink]

Reopening till #515 is merged

[stefvanbuuren]

Ah, thanks for the diagnosis. Let's first solve #515 then.