tmp_index is shorter than to_index sometimes, especially in small and degenerate blockmodels

[alexpghayes]

Playing around with different seeds in the degenerate SBM test reveals a variety of new and exciting issues. TL; DR: if you need to sample small graphs (or potentially SBMs with k ~ n), you may run into issues.

For some reason this errors for me when run by testthat but not interactively.

library(fastRG)

set.seed(5)

n <- 10
k <- 10
pi <- rep(1, k) / k

B <- diag(rep(0.5, k))

sbm <- sbm(n = n, pi = pi, B = B)

sample_sparse(sbm)

If this becomes a blocker for anyone, please let me know.

[alexpghayes]

Reprex

library(fastRG)
#> Loading required package: Matrix

set.seed(7)

n <- 10
k <- 10
pi <- rep(1, k) / k

B <- diag(rep(0.5, k))

sbm <- sbm(n = n, pi = pi, B = B)

sample_sparse(sbm)
#> Error in to[to_index] <- to_tmp[tmp_index]: replacement has length zero

^{Created on 2022-06-29 by the reprex package (v2.0.1)}

Can also get a version of this that is only a warning:

> sample_sparse(sbm)
10 x 10 sparse Matrix of class "dsCMatrix"

[1,] . . . . . . . . . .
[2,] . . . . . . . . . .
[3,] . . . . . . . . . .
[4,] . . . . 3 1 . . . .
[5,] . . . 3 1 . . . . .
[6,] . . . 1 . . . . . .
[7,] . . . . . . . . . .
[8,] . . . . . . . . . .
[9,] . . . . . . . . . .
[10,] . . . . . . . . . .
Warning message:
  In to[to_index] <- to_tmp[tmp_index] :
  number of items to replace is not a multiple of replacement length

[alexpghayes]

@karlrohe I'm pretty sure this I found this fix for this, but if you have a moment and you can take a look at e9ea60e, I would appreciate it! Only a one line change of code for what I think is the fix.