Remove default seed from runNMF

[jashapiro]

When working with the calculateNMF() function (with the release version of scater), I realized that the default arguments result in resetting the system seed, which seems like undesired behavior, both because it means that repeated runs do not vary as would be expected (and as the docs seem to indicate).

For example:

sce <- mockSCE() |> logNormCounts()

set.seed(100)
nmf1 <- calculateNMF(sce)
runif(1)
#> [1] 0.5308088
nmf2 <- calculateNMF(sce)
runif(1)
#> [1] 0.5308088
all(nmf1==nmf2)
#> [1] TRUE

Setting the default seed as NULL seems to correctly alleviate the issue (and this is the default seed value for the underlying RccpML::nmf()):

set.seed(100)
nmf1 <- calculateNMF(sce, seed = NULL)
runif(1)
#> [1] 0.0740483

# second run is different and not in the same RNG position
nmf2 <- calculateNMF(sce, seed = NULL)
runif(1)
#> [1] 0.7676133
all(nmf1==nmf2)
#> [1] FALSE

# resetting the seed does result in identical values, as expected
set.seed(100)
nmf3 <- calculateNMF(sce, seed = NULL)
runif(1)
#> [1] 0.0740483
all(nmf1==nmf3)
#> [1] TRUE

^{Created on 2025-02-28 with reprex v2.1.1}

Since the pattern from other similar functions like calculateUMAP() is to not include seed as a default argument, the simplest solution seemed to me to be to simply remove the argument, allowing the default value for RcppML::nmf() to be used. That is what I have done here.

I also fixed a small error in the docs which referred to passing additional arguments to Rtsne.

If you would like to pursue a different solution (or leave it as is), please let me know.

[alanocallaghan]

Looks sensible, I have no idea why I exposed the seed argument in particular

[alanocallaghan]

Thanks for this!

[LTLA]

Looks sensible, I have no idea why I exposed the seed argument in particular

Might have been me. But I too have no idea why that was done.