2017/12/22 10:40-11:30 Survey of Current Random Number Generators used in R and Comparison with DX Generators

TitleSurvey of Current Random Number Generators used in R and Comparison with DX Generators

Speaker:Prof. Lih-Yuan Deng (University of Memphis, USA)

Time:106/12/22/(Fri.)  10:40-11:30

Location:Room 427, Assembly Building, NCTU

Abstract

The R software system has become increasingly popular and as such it is interesting to examine the PRNGs available for use in the base package of R. In the base package the random number generating function (RNGkind) has options for specifying a particular PRNG that includes

(1) “Mersenne-Twister”, the MT19937 generator,

(2)  “Wichmann-Hill”, a combination generator,

(3) “Marsaglia-Multicarry”, a multiply-with-carry generator,

(4) “Super-Duper”  combining LCG and LFSR as proposed by  Marsaglia,

(5) “Knuth-TAOCP-2002”, a lagged Fibonacci generator proposed by Knuth,

(6) “Knuth-TAOCP” an older version of  “Knuth-TAOCP-2002” with a different initialization, and

(7) “L’Ecuyer-CMRG” (MRG32k3a), a combination generator combining two MRGs of order 3, proposed by L’Ecuyer.

In this talk, we review and compare these PRNGs available in RNGkind and compare them with the DX(Deng-Xu) generator, a fast and efficient, huge period multiple recursive generator (MRG) with equi-distribution in thousands of dimensions. Comparing with these generators, DX generators are fastest with longest period length.   Furthermore, we show that DX generators consistently pass all the empirical tests while most of the PRNGs used in base R are not able to pass all these tests, including the default generator MT19937.

 

Organizer:NCTU Big Data Research Center

Co-organiser:NCTU Institute of Statistics、NTHU Institute of Statistics