R/random_numbers.R

In interpret: Fit Interpretable Machine Learning Models

# Copyright (c) 2018 Microsoft Corporation
# Licensed under the MIT license.
# Author: Paul Koch <code@koch.ninja>

normalize_initial_seed <- function(seed) {
   # Some languages do not support 64-bit values.  Other languages do not support unsigned integers.
   # Almost all languages support signed 32-bit integers, so we standardize on that for our 
   # random number seed values.  If the caller passes us a number that doesn't fit into a 
   # 32-bit signed integer, we convert it.  This conversion doesn't need to generate completely 
   # uniform results provided they are reasonably uniform, since this is just the seed.
   # 
   # We use a simple conversion because we use the same method in multiple languages, 
   # and we need to keep the results identical between them, so simplicity is key.
   # 
   # The result of the modulo operator is not standardized accross languages for 
   # negative numbers, so take the negative before the modulo if the number is negative.
   # https://torstencurdt.com/tech/posts/modulo-of-negative-numbers

   if(is.null(seed)) {
      return(NULL)
   }

   if(2147483647 <= seed) {
      return(seed %% 2147483647)
   }
   if(seed <= -2147483647) {
      return(-((-seed) %% 2147483647))
   }
   return(seed)
}

create_rng <- function(random_state) {
   if(is.null(random_state)) {
      return(NULL)
   }
   random_state <- as.integer(random_state)

   stopifnot(-2147483648 <= random_state && random_state <= 2147483647)

   ret <- .Call(CreateRNG_R, random_state)
   return(ret)
}