CODEWITHSUNDEEP
c++randomrcpp
Benno Bartels
Benno Bartels

Reputation: 21

Set c++ rng seed from r using cpp

I am writing a class in c++ that uses a c++ rng to produce random numbers. The constructor takes the seed for the rng as an argument. However, when I expose the class to r using the rcpp package setting the seed does not work as I get different random numbers every time I instantiate a member of that class even though I pass the same seed. Is there a trick how to set the seed in c++ when using rcpp? Happy for all the help I can get.

Upvotes: 0

Views: 774

Answers (2)

Benno Bartels
Benno Bartels

Reputation: 21

Thanks for all the help. I found the problem. I was using a C++ RNG and initialised the generator as a static variable in a member function without making it an explicit member of the class. It works fine in C++, but does not seem to work when the code is called from R for some Reason. To clarify: 

//.h file
class RandomGen{
private:
   const unsigned int seed;
public:
   RandomGen(unsigned int _seed);
   double get_num();
};

//.cpp file
RandomGen::RandomGen(unsigned int _seed):
seed(_seed)
{}

double RandomGen::get_num(){
    static std::mt19937 generator(seed);
    static std::exponential_distribution<double> distrMarket(1);
    return distrMarket(generator);
}

Reproduces the problem. Making the rng generator a static member variable of the class fixes the problem. I just have no idea why.

Upvotes: 0

Dirk is no longer here
Dirk is no longer here

Reputation: 368211

Which RNG are you trying to seed? If it is R's you are more or less out of luck as Writing R Extensions clearly says you should do it from R.

There are workarounds (eg via Rcpp::Function()) but maybe you should rethink the problem.

If however your are seeding your own RNG, it should work, and works eg for my RcppZiggurat package:

R> library(RcppZiggurat)      
R> zsetseedLZLLV(12345)                 # set seed
R> zrnormLZLLV(3)
[1]  1.297770 -0.699053  0.443079
R> zsetseedLZLLV(12345)                 # re-set seed
R> zrnormLZLLV(3)
[1]  1.297770 -0.699053  0.443079       # same number
R> zrnormLZLLV(3)
[1] -0.405296 -1.771543  0.319896       # diff. now
R>

Lastly, if your question about to make your RNG be used by R's functions: that is possible (again, see Writing R Extensions) but maybe not such a great idea as code calling runif() or rnorm() has some (implicit) expectactions about what would get called.

Upvotes: 1

Related Questions