How to set seed when using `boot::boot |> futurize()`?

[jarbet]

When parallelizing the bootstrap with futurize() by doing boot::boot(...) |> futurize(), how should I properly set the seed for reproducibility?

My past experience with the futureverse is using future.apply::future_lapply() which has an argument future.seed that can be used to control reproducibility. Not sure what the best practice is for setting the seed with futurize?

The methods I've tried below fail to properly set the seed:

futurize_options(seed = 123)

suppressPackageStartupMessages({
    library(future)
    library(futurize)
    library(boot)
})

future::plan(future::multisession, workers = 5)
futurize_options(seed = 123)
#> $seed
#> [1] 123
#> 
#> $globals
#> [1] TRUE
#> 
#> $packages
#> NULL
#> 
#> $stdout
#> [1] TRUE
#> 
#> $conditions
#> [1] "condition"
#> 
#> $scheduling
#> [1] 1
#> 
#> $chunk_size
#> NULL
#> 
#> attr(,"specified")
#> [1] "seed"

ratio <- function(d, w) sum(d$x * w) / sum(d$u * w)
boot(city, ratio, R = 100, stype = "w") |> futurize()
#> 
#> ORDINARY NONPARAMETRIC BOOTSTRAP
#> 
#> 
#> Call:
#> boot(data = city, statistic = ratio, R = 100, stype = "w", parallel = "snow", 
#>     ncpus = 2, cl = cl)
#> 
#> 
#> Bootstrap Statistics :
#>     original     bias    std. error
#> t1* 1.520312 0.02174888   0.2072925
boot(city, ratio, R = 100, stype = "w") |> futurize()
#> 
#> ORDINARY NONPARAMETRIC BOOTSTRAP
#> 
#> 
#> Call:
#> boot(data = city, statistic = ratio, R = 100, stype = "w", parallel = "snow", 
#>     ncpus = 2, cl = cl)
#> 
#> 
#> Bootstrap Statistics :
#>     original     bias    std. error
#> t1* 1.520312 0.05856933   0.2216266

futurize(seed = 123)

suppressPackageStartupMessages({
    library(future)
    library(futurize)
    library(boot)
})

future::plan(future::multisession, workers = 5)

ratio <- function(d, w) sum(d$x * w) / sum(d$u * w)
boot(city, ratio, R = 100, stype = "w") |> futurize(seed = 123)
#> 
#> ORDINARY NONPARAMETRIC BOOTSTRAP
#> 
#> 
#> Call:
#> boot(data = city, statistic = ratio, R = 100, stype = "w", parallel = "snow", 
#>     ncpus = 2, cl = cl)
#> 
#> 
#> Bootstrap Statistics :
#>     original       bias    std. error
#> t1* 1.520312 -0.002385279   0.1700565
boot(city, ratio, R = 100, stype = "w") |> futurize(seed = 123)
#> 
#> ORDINARY NONPARAMETRIC BOOTSTRAP
#> 
#> 
#> Call:
#> boot(data = city, statistic = ratio, R = 100, stype = "w", parallel = "snow", 
#>     ncpus = 2, cl = cl)
#> 
#> 
#> Bootstrap Statistics :
#>     original     bias    std. error
#> t1* 1.520312 0.03643624   0.1942623

[HenrikBengtsson]

Hi. The following works:

set.seed(123)
b1 <- boot::boot(boot::city, ratio, R = 999) |> futurize(seed = TRUE)

set.seed(123)
b2 <- boot::boot(boot::city, ratio, R = 999) |> futurize(seed = TRUE)

all.equal(b1, b2)
#> [1] TRUE

Turns out seed = <integer> is not supported by boot. I might be able to work around that internally in futurize().

That said, do you really need to specify seed = <integer>?

PS. futurize_options() is just a convenient tool for generating a list of future options - but it does not apply any settings. So, just use futurize(...).

[HenrikBengtsson]

As a starter, next version will detect this and give an informative error message, e.g.

> b <- boot::boot(boot::city, ratio, R = 999) |> futurize(seed = 123)
Error in transpile(expr, options = options) : 
  futurize() option 'seed' must be logical for this function call, not numeric: boot::boot(boot::city, ratio, R = 999)