| 1 |
# -*- org -*- |
| 2 |
#+CATEGORY: rng |
| 3 |
|
| 4 |
* gfsr: for consistency (Charles A. Wemple). Make this change in 2.0 |
| 5 |
|
| 6 |
127a142 |
| 7 |
> state->nd = i-1; |
| 8 |
133,134c148,149 |
| 9 |
< for (i=0; i<32; ++i) { |
| 10 |
< int k=7+i*3; |
| 11 |
--- |
| 12 |
> for (i=0; i<32; i++) { |
| 13 |
> int k=7*i+3; |
| 14 |
141d155 |
| 15 |
< state->nd = i; |
| 16 |
|
| 17 |
|
| 18 |
* Sort out "const" in prototypes, it looks odd since there are consts |
| 19 |
for many functions which modify the state. (Applies to both qrng and rng) |
| 20 |
|
| 21 |
* New 64 bit generators in "Tables of 64-bit Mersenne Twisters", |
| 22 |
Takuji Nishimura, ACM Transactions on Modeling and Computer |
| 23 |
Simulation, Volumen 10, No 4, October 2000, p 348--357 |
| 24 |
|
| 25 |
* Need to run tests over the space of seeds, in addition to serial |
| 26 |
tests (DIEHARD) for the default seed. The congruences used for |
| 27 |
seeding will give poor initial vectors for some seed values, e.g. 2^n. |
| 28 |
Improve the seeding procedure by using a high-quality generator |
| 29 |
(e.g. hash functions, md5 or sha) to generate the initial vectors. |
| 30 |
Even if this is moderately expensive it is worthwhile since the |
| 31 |
seeding is usually a one-off cost at program initialization, and the |
| 32 |
results of all subsequent calls to the generator depend on it. The |
| 33 |
GSFR4 generator is particularly likely to benefit from this procedure. |
| 34 |
|
| 35 |
* Add SWNS generator Phys.Rev.E 50 (2) p. 1607-1615 (1994), Phys.Rev.E |
| 36 |
60 (6), p.7626-7628 (1999) |
| 37 |
|
| 38 |
* Add get_array type methods which can provide optimized versions of |
| 39 |
each function (?). We should offer the possibility of eliminating |
| 40 |
function call overhead -- there are various possible ways to implement |
| 41 |
it. |
| 42 |
|
| 43 |
* Add ISAAC generator (??) |
| 44 |
|
| 45 |
* Add (A)RC4 and hash based random number generators MD5, SHA. This |
| 46 |
should give crypto quality randomness, and guarantee different |
| 47 |
sequences for each seed. Make the state (seed, count) where count |
| 48 |
increments on each call of the generator. Implement count as a big |
| 49 |
integer stored in separate unsigned integers so that the period is |
| 50 |
sufficiently long (e.g. 2^64 or 2^96). The generator would return |
| 51 |
HASH(seed, count) on each call. |
| 52 |
|
| 53 |
* Check that RANLXS will work on machines with non-standard width of |
| 54 |
float/dbl (original has checks for DBL_MANT_DIG ..) |
| 55 |
|
| 56 |
* mention more clearly why not all Cokus used (or recheck MT pages for |
| 57 |
improvements) |
| 58 |
|
| 59 |
* run the DIEHARD tests on all the generators, especially the ones we |
| 60 |
are listing as "Simulation Quality" -- some of those are a bit old and |
| 61 |
might fail one or two diehard tests. |
| 62 |
|
| 63 |
* Add NAG, missing, gave up! |
| 64 |
CDC 48-bit missing |
| 65 |
|
| 66 |
* Check out the bug fix to mrand48 that was made in glibc2, pr757 |
| 67 |
|
| 68 |
* Check out the following paper, |
| 69 |
|
| 70 |
On the anomaly of ran1() in Monte Carlo pricing of financial |
| 71 |
derivatives; Akira Tajima , Syoiti Ninomiya , and Shu Tezuka ; Winter |
| 72 |
simulation , 1996, Page 360, from ACM |
| 73 |
|
| 74 |
* The following papers have been published, I think we refer to them |
| 75 |
(or could do), |
| 76 |
|
| 77 |
Pierre L'Ecuyer, "Tables of Linear Congruential Generators of different |
| 78 |
size and good lattice structure", Mathematics of Computation, Vol 68, |
| 79 |
No 225, Jan 1999, p249-260 |
| 80 |
|
| 81 |
Pierre L'Ecuyer, "Tables of Maximally equidistributed combined LSFR |
| 82 |
generators", ibid, p261-270 |
| 83 |
|
| 84 |
* Look at this paper: I. Vattulainen, "Framework for testing random numbers |
| 85 |
in parallel calculations", Phys. Rev. E, 59, 6, June 1999, p7200 |
| 86 |
|
| 87 |
---------------------------------------------------------------------- |
| 88 |
DONE |
| 89 |
|
| 90 |
x1. Improve the seeding for routines that use the LCG seed generator. |
| 91 |
It can only generate 130,000 different initial states. We should |
| 92 |
change it to provide 2^31 different initial states (this will also |
| 93 |
prevent the high bits being zero). DONE - we now use a 32-bit |
| 94 |
generator. |
| 95 |
|
| 96 |
x8. Get the macros from the faster MT19937 generator and use them. We |
| 97 |
need to make MT be the fastest of the simulation quality generators if |
| 98 |
it is the default. DONE. It didn't improve the speed on other |
| 99 |
platforms, so I just used the tricks which also worked on the pentium |
| 100 |
(e.g. changing mag[x&1] to x&1 ? mag[1] : mag[0]) |