tango.math.random.Random

Random number generators

This is an attempt at having a good flexible and easy to use random number generator. ease of use:

• shared generator for quick usage available through the "rand" object

rand.distributionD!(type)(paramForDistribution)

i=r.uniformR2(5,10); // uniform numbers in (5;10) f=r.uniformR2(5.0f,10.0f); rv=r.uniformR2(5.0L,10.0L); foreach (ref el;ar) el=r.uniformR2(5.0L,10.0L); // another way to do all the previous in one go: r.uniformR2D(5.0L,10.0L)(i)(f)(r)(ar); // uniform distribution -10..10 i=r.uniformRSymm(10); // well you get it... r.uniformRSymmD(10)(i)(f)(r)(ar); // any distribution can be stored auto r2=r.uniformRSymmD(10); // and used later r2(ar); // complex distributions (normal,exp,gamma) are produced for the requested type r.normalSource!(float)()(f); // with sigma=2 r.normalD(2.0f)(f); // and can be used also to initialize other types r.normalSource!(float)()(r)(ar); r.normalD(2.0f)(r)(ar); // but this is different from r.normalSource!(real)()(i)(r)(ar); r.normalD(2.0L)(i)(r)(ar); // as the source generates numbers of its type that then are simply cast to // the type needed. // Uniform distribution (as its creation for different types has no overhead) // is never cast, so that (for example) bounds exclusion for floats is really // guaranteed. // For the other distribution using a distribution of different type than // the variable should be done with care, as underflow/overflow might ensue. // // Some utility functions are also available int i2=r.uniform!(int)(); int i2=r.randomize(i); // both i and i2 are initialized to the same value float f2=r.normalSigma(3.0f);

  )
)
flexibility:
$(UL
  $(LI  easily swappable basic source

// a random generator that uses the system provided random generator: auto r=RandomG!(Urandom)();

1         One could also build an engine that can be changed at runtime (that calls
2         a delegate for example), but this adds a little overhead, and changing
3         engine is not something done often, so this is not part of the library.
4   )
5   $(LI  ziggurat generator can be easily adapted to any decreasing derivable
6         distribution, the hard parametrization (to find xLast) can be done
7         automatically
8   )
9   $(LI  several distributions available "out of the box"
10   )
11   )
12   Quality:
13   $(UL
14   $(LI  the default Source combines two surces that pass all statistical tests 
15         (KISS+CMWC)
16         (P. Lcuyer and R. Simard, ACM Transactions on Mathematical Software (2007),
17         33, 4, Article 22, for KISS, see CMWC engine for the other)
18   )
19   $(LI  floating point uniform generator always initializes the full mantissa, the
20         only flaw is a (*very* small) predilection of 0 as least important bit 
21         (IEEE rounds to 0 in case of tie).
22         Using a method that initializes the full mantissa was shown to improve the
23         quality of subsequntly derived normal distribued numbers
24         (Thomas et al. Gaussian random number generators. Acm Comput Surv (2007)
25         vol. 39 (4) pp. 11))
26   )
27   $(LI  Ziggurat method, a very fast and accurate method was used for both Normal and
28         exp distributed numbers.
29   )
30   $(LI  gamma distribued numbers uses a method recently proposed by Marsaglia and
31         Tsang. The method is very fast, and should be good.
32         My (Fawzi) feeling is that the transformation h(x)=(1+d*x)^3 might lose
33         a couple of bits of precision in some cases, but it is unclear if this
34         might become visible in (*very* extensive) tests or not.
35   )
36    the basic source can be easily be changed with something else
37   Efficiency:
38   $(LI  very fast methods have been used, and some effort has been put into
39         optimizing some of them, but not all, but the interface has been choosen
40         so that close to optimal implementation can be provided through the same
41         interface.
42   )
43   $(LI  Normal and Exp sources allocated only upon request: no memory waste, but
44         a (*very* small) speed hit, that can be avoided by storing the source in
45         a variable and using it (not going through the RandomG)
46   )
47 )
48 Annoyances:
49 $(UL
50   $(LI  I have added two "next" methods to RandomG for backward compatibility
51         reasons, and the .instance from Random has been
52         replaced by the "rand" object. The idea behind this is that RandomG is
53         a template and rand it should be shared across all templates.
54         If the name rand is considered bad one could change it. 
55         I kept .instance static method that returns rand, so this remain a dropin
56         replacement of the old random.
57   )
58   $(LI You cannot initialize a static array directly, this because randomize is
59       declared like this:

U randomize(U)(ref U a) { }

make arrays initialized, and scalar not, which is much worse. ) )