18 Random number generation
R’s RNG is private to R: from C you draw from R’s stream, you don’t create your own. The stream state lives in the .Random.seed variable in the global environment, so every use from C must be bracketed by GetRNGstate() (read the state in, creating it if needed) and PutRNGstate() (write it back). Skip the bracket and your draws won’t advance R’s stream — repeated calls return the same values, and set.seed() at the R level won’t make your code reproducible.
There is no C API to choose the generator or set the seed; if you need that, evaluate calls to RNGkind()/set.seed() (see Evaluation). For distributions beyond uniform, normal, and exponential, use the r* functions from Rmath.h (see Mathematical functions) — they draw from the same stream and need the same bracketing.
18.0.1 GetRNGstate(), PutRNGstate()
throws
Header: R_ext/Random.h
Read in (or create) .Random.seed before generating variates, and write it back after.
void GetRNGstate(void);
void PutRNGstate(void);Every use of R’s RNG from C must be bracketed by these calls: GetRNGstate() reads (or lazily creates) .Random.seed, and PutRNGstate() writes the updated state back so R-level code sees the stream advance. They can allocate when the seed doesn’t yet exist. These names are never remapped.
GetRNGstate();
double u = unif_rand();
PutRNGstate();See also: unif_rand()
18.0.2 unif_rand(), norm_rand(), exp_rand()
Header: R_ext/Random.h
Draw one uniform, standard normal, or unit exponential variate.
double unif_rand(void);
double norm_rand(void);
double exp_rand(void);Returns: One variate: uniform on (0, 1) for unif_rand(), standard normal for norm_rand(), unit exponential for exp_rand().
Must be called between GetRNGstate() and PutRNGstate(). The generator kind and seed are chosen at the R level (RNGkind(), set.seed()); there is no C API to select them. For other distributions, use the r* functions from Rmath.h (see Mathematical functions) — they need the same bracketing. These names are never remapped.
See also: GetRNGstate(), R_unif_index()
18.0.3 R_unif_index()
Header: R_ext/Random.h
Draw a uniform index in [0, n) for random sampling.
double R_unif_index(double n);n: upper bound (exclusive) on the returned index.
Returns: A uniform draw in [0, n); truncate to an integer type to get an index.
Returns a double in [0, n); truncate to int or R_xlen_t to get an index. Uses the sampling algorithm selected by R’s sample.kind option, so results match R-level sample.int(). Must be called between GetRNGstate() and PutRNGstate().
See also: unif_rand()
18.0.4 R_sample_kind()
Header: R_ext/Random.h
Get the current discrete-sampling algorithm.
Sampletype R_sample_kind(void);Returns: The current sampling algorithm, ROUNDING or REJECTION.
Returns ROUNDING or REJECTION, reflecting the sample.kind option; only relevant to code implementing its own discrete sampling.
See also: R_unif_index()