4

I am debugging a larger set of nested models that only run into problems during optimization. During the process of zeroing on what I believe is causing the errors I've come across unusual behavior in the rpois() function.

It seems that with very large mean values, rpois() returns NA instead of a value. This problem does not generate a warning. See below for a reproducible set of code. 

> rpois(1,3000000000)
[1] NA

My question is two fold:
1 - why is it showing this behavior (is there a max limit on the size of an integer for the rpois function?) and
2 - is there a work around to prevent the generation of NA (even if that's to limit the size of the mean input to some smaller value)?

I am running 32x R version 3.0.2 in 64x Windows 7.

thelatemail
  • 91,185
  • 12
  • 128
  • 188
user3658540
  • 41
  • 2

1 Answers1

7

The problem is that rpois returns an integer, and it converts the value to NA if the value is greater than the maximum possible integer value (.Machine$integer.max).

rpois(1,.Machine$integer.max/1.00001)
## [1] 2147428954
rpois(1,.Machine$integer.max/1)
## [1] NA

The Normal approximation should be insanely precise in this case (it's generally extremely good if the mean is greater than 100!): if your mean is greater than (say) 0.999*.Machine$integer.max, you can use round(rnorm(1,mean=lambda,sd=sqrt(lambda)))

Ben Bolker
  • 211,554
  • 25
  • 370
  • 453