Performance: Its speed is comparable to other modern generators (in particular, as fast as java.util.Random.nextFloat()). 2.5 million calls to raw() per second (Pentium Pro 200 Mhz, JDK 1.2, NT). Be aware, however, that there is a non-negligible amount of overhead required to initialize the data structures used by a MersenneTwister. Code like
double sum = 0.0; for (int i=0; i<100000; ++i) { RandomElement twister = new MersenneTwister(new java.util.Date()); sum += twister.raw(); } will be wildly inefficient. Consider using double sum = 0.0; RandomElement twister = new MersenneTwister(new java.util.Date()); for (int i=0; i<100000; ++i) { sum += twister.raw(); } instead. This allows the cost of constructing the MersenneTwister object to be borne only once, rather than once for each iteration in the loop. Implementation: After M. Matsumoto and T. Nishimura, "Mersenne Twister: A 623-Dimensionally Equidistributed Uniform Pseudo-Random Number Generator", ACM Transactions on Modeling and Computer Simulation, Vol. 8, No. 1, January 1998, pp 3--30.
The correctness of this implementation has been verified against the published output sequence mt19937-2.out of the C-implementation mt19937-2.c. (Call test(1000) to print the sequence).
Details: MersenneTwister is designed with consideration of the flaws of various existing generators in mind. It is an improved version of TT800, a very successful generator. MersenneTwister is based on linear recurrences modulo 2. Such generators are very fast, have extremely long periods, and appear quite robust. MersenneTwister produces 32-bit numbers, and every k-dimensional vector of such numbers appears the same number of times as k successive values over the period length, for each k <= 623 (except for the zero vector, which appears one time less). If one looks at only the first n <= 16 bits of each number, then the property holds for even larger k, as shown in the following table (taken from the publication cited above):
| n | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 .. 16 | 17 .. 32 |
| k | 19937 | 9968 | 6240 | 4984 | 3738 | 3115 | 2493 | 2492 | 1869 | 1869 | 1248 | 1246 | 623 |
MersenneTwister generates random numbers in batches of 624 numbers at a time, so the caching and pipelining of modern systems is exploited. The generator is implemented to generate the output by using the fastest arithmetic operations only: 32-bit additions and bit operations (no division, no multiplication, no mod). These operations generate sequences of 32 random bits (int's). long's are formed by concatenating two 32 bit int's. float's are formed by dividing the interval [0.0,1.0] into 232 sub intervals, then randomly choosing one subinterval. double's are formed by dividing the interval [0.0,1.0] into 264 sub intervals, then randomly choosing one subinterval.
@author wolfgang.hoschek@cern.ch @version 1.0, 09/24/99 @see java.util.Random
| |
| |
| |
| |