Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

The native VB Rnd() function can generate a sequence of upto only 16,777,216 numbers before they are repeated. This is adequate in most cases but some projects may require many more numbers.

Below is what I wrote in Excel (can be used in VB except the TestSub) based on the algorithm developed by B.A. Wichman and I.D. Hill that is implemented in Excel 2003 for WorksheetFunction RAND().

As said, this algorithm guarantees to provide a sequence of more than 10,000,000,000,000 (10^13) numbers before they are repeated.

:) I am happy with everyone's input to make this function better and faster.

:sick: Please point out any problem you see.

;) On my PC, withExcel 2003, it can generate 1 million numbers in every around 550 milliseconds (including time for storing values.)

Put the top 3 Code blocks in one module and the test Sub in another module.
Have fun!:)

Code:

'==================================================================================== 'http://support.microsoft.com/kb/828795 ' 'The basic idea is that if you take three random numbers on [0,1] and sum them, 'the fractional part of the sum is itself a random number on [0,1]. 'The critical statements in the Fortran code listing from the original 'Wichman and Hill article are: ' 'C IX, IY, IZ SHOULD BE SET TO INTEGER VALUES BETWEEN 1 AND 30000 BEFORE FIRST ENTRY 'IX = MOD(171 * IX, 30269) 'IY = MOD(172 * IY, 30307) 'IZ = MOD(170 * IZ, 30323) 'RANDOM = AMOD(FLOAT(IX) / 30269.0 + FLOAT(IY) / 30307.0 + FLOAT(IZ) / 30323.0, 1.0) '==================================================================================== Option Explicit Private ix As Long, iy As Long, iz As Long

Code:

Sub RandomizeX(Optional ByVal Number) Const MaxLong As Double = 2 ^ 31 - 1 Dim n As Long Dim d As Double If IsMissing(Number) Then n = Timer * 60 '-- Timer is only updated in every 1/60th second. '-- Multiply by 60 to reduce the chance of seed '-- to be repeated in subsequence calls of RandomizedX Else d = Abs(Int(Val(Number))) If d > MaxLong Then '-- prevent Long overflow d = d - Int(d / MaxLong) * MaxLong End If n = d End If ix = (n Mod 30269) iy = (n Mod 30307) iz = (n Mod 30323) '-- ix, iy, iz cannot be 0 If ix = 0 Then ix = 171 If iy = 0 Then ix = 172 If iz = 0 Then ix = 170 End Sub

Code:

Function RndX(Optional ByVal Number As Long = 1) As Double Dim r As Double If ix = 0 Then '-- ix, iy, iz cannot be 0. ix = 171 '-- Initial values of ix, iy and iz are 0. iy = 172 '-- If any of these equal 0, iz = 170 '-- it will be stucked with 0 forever. End If If Number <> 0 Then If Number < 0 Then RandomizeX Number End If ix = (171 * ix) Mod 30269 '-- This has been tested: iy = (172 * iy) Mod 30307 '-- ix, iy, iz will never be 0. iz = (170 * iz) Mod 30323 '-- --------------------------- End If r = ix / 30269# + iy / 30307# + iz / 30323# RndX = r - Int(r) End Function

Run TestRndX(). If you want to stop, use Ctrl-Break then un-remark 'Exit Do
then let the Sub continues to exit properly to reset ScreenUpdating.

Code:

Option Explicit Private Declare Function GetTickCount Lib "kernel32" () As Long Sub TestRndX() Dim r0 As Double Dim r As Double Dim rMin As Double Dim rMax As Double Dim rSum As Double Dim rAvg As Double Dim TenPctCount(0 To 9) As Double Dim p As Integer Dim n As Long Dim m As Long Dim t As Double Dim aCell As Range Dim CellRow As Long Const BottomRow = 65536 Const Million = 1000000 Dim Headers As Variant Dim ColWidths As Variant '----- Prepare Sheet for report --------------------------------------- Headers = Array("Count", "Cur Rnd", "millisecs", "Min", "Max", "Avg", _ "<0.1", "<0.2", "<0.3", "<0.4", "<0.5", _ "<0.6", "<0.7", "<0.8", "<0.9", "<1.0") ColWidths = Array(15, 18, 8, 18, 18, 18, _ 13, 13, 13, 13, 13, 13, 13, 13, 13, 13) With Sheet1 For p = 0 To UBound(Headers) .Cells(1, p + 1).ColumnWidth = ColWidths(p) Next .Range(.Cells(1, 1), .Cells(1, 16)) = Headers .Range("A:A,C:C,G:P").NumberFormat = "#,##0" .Range("B:B,D:F").NumberFormat = "0.000000000000000" '----------------------------------------------------------------------- RandomizeX r0 = RndX '---- FIRST Random Number rMin = r0 rMax = r0 rSum = r0 rAvg = r0 CellRow = 2 .Cells(CellRow, 1) = 0 .Cells(CellRow, 2) = r0 Application.ScreenUpdating = False Do While CellRow < BottomRow '========================= 'Exit Do '-- Un-remark this after Control-Break then let ' the Sub continue to reset ScreenUpdating '========================= t = GetTickCount For n = 1 To Million r = RndX If rMin > r Then rMin = r If rMax < r Then rMax = r rSum = rSum + r p = Int(r * 10) TenPctCount(p) = TenPctCount(p) + 1 If r = r0 Then .Cells(CellRow, 1) = m * Million + n .Cells(CellRow, 2) = r .Cells(CellRow, 3) = "Repeated!" Exit Do End If Next t = GetTickCount - t m = m + 1 CellRow = CellRow + 1 .Cells(CellRow, 1) = CDbl(m) * Million .Cells(CellRow, 2) = r .Cells(CellRow, 3) = t .Cells(CellRow, 4) = rMin .Cells(CellRow, 5) = rMax .Cells(CellRow, 6) = rSum / (CDbl(m) * Million) '-- rAvg For p = 0 To 9 .Cells(CellRow, 7 + p) = TenPctCount(p) Next If m Mod 10 = 0 Then Application.ScreenUpdating = True ActiveWindow.ScrollRow = CellRow - 10 DoEvents Application.ScreenUpdating = False End If Loop End With Application.ScreenUpdating = True End Sub

Re: Wichman-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

Here is a suggestion for independent function. It doesn't work exactly the same, but I paid a lot attention to details on how it works.

Code:

Public Function RndM(Optional ByVal Number As Long) As Double Static lngX As Long, lngY As Long, lngZ As Long, blnInit As Boolean Dim dblRnd As Double ' if initialized and no input number given If blnInit And Number = 0 Then ' lngX, lngY and lngZ will never be 0 lngX = (171 * lngX) Mod 30269 lngY = (172 * lngY) Mod 30307 lngZ = (170 * lngZ) Mod 30323 Else ' if no initialization, use Timer, otherwise ensure positive Number If Number = 0 Then Number = Timer * 60 Else Number = Number And &H7FFFFFFF lngX = (Number Mod 30269) lngY = (Number Mod 30307) lngZ = (Number Mod 30323) ' lngX, lngY and lngZ must be bigger than 0 If lngX > 0 Then Else lngX = 171 If lngY > 0 Then Else lngY = 172 If lngZ > 0 Then Else lngZ = 170 ' mark initialization state blnInit = True End If ' generate a random number dblRnd = CDbl(lngX) / 30269# + CDbl(lngY) / 30307# + CDbl(lngZ) / 30323# ' return a value between 0 and 1 RndM = dblRnd - Int(dblRnd) End Function

Here are a few pointers:
True conditions are faster than False conditions -> set code that is ran more often behind True condition.
I minimized the amount of calculation when checking for initialized values. By using a local static variable we know that X, Y and Z are valid.
I decided to use 0 as number to cause Timer to be used automatically for seeding: if someone wants repeatable random numbers, he knows to give a seed.
... which leads to this function having an advantage over Rnd as you don't need to call Randomize to get a different set of random numbers on each run of the application.
And I used variable names that tell better what datatype they represent.

Re: Wichman-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

Quote:

Originally Posted by Merri

Here are a few pointers:
True conditions are faster than False conditions -> set code that is ran more often behind True condition.
I minimized the amount of calculation when checking for initialized values. By using a local static variable we know that X, Y and Z are valid.
I decided to use 0 as number to cause Timer to be used automatically for seeding: if someone wants repeatable random numbers, he knows to give a seed.
... which leads to this function having an advantage over Rnd as you don't need to call Randomize to get a different set of random numbers on each run of the application.
And I used variable names that tell better what datatype they represent.

@Merri,
I agree with what you mentioned. At first I wrote a single function with Static vars as you did without Randomize sub but later I changed my mind to mimic the existing pair Rnd() and Randomize() so we don't need to explain how to use it.
But I like your way as it look simpler. However your function won't run faster than my version, if not about 5% slower. The problem makes it run slower is Static vars (I am not sure what behind the scenes.)
If all vars are declared as Module Private vars instead of Function Static vars, it will boost up the speed about 20%, that means it will be faster than my version about 15%.

Code:

Option Explicit Private lngX As Long, lngY As Long, lngZ As Long, blnInit As Boolean Public Function RndM(Optional ByVal Number As Long) As Double '--Static lngX As Long, lngY As Long, lngZ As Long, blnInit As Boolean Dim dblRnd As Double ... ...

1 Attachment(s)

Re: Wichman-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

I ran a test on both functions. I used the return value to fill a big 32Mb byte array something like this snippet...

Code:

i = Int(RndX * &H2000000) If RndResult(i) = 255 Then Exit For Else RndResult(i) = RndResult(i) + 1 End If

...in order to count all the fractions (0/33554432, 1/33554432, 2/33554432 .... 33554431/33554432) that function would return. It stopped counting when 1 fraction was returned 255 times, it then dumped the data to a file.

In the image below each horizontal pixel represents 32768 readings, the vertical pixels show the exact count. Around 6,000,000,000 readings were made. If the function was biased it would not show a straight line. Both functions generate nearly identical images. Very random, clear mean and variance and no obvious bias. :thumb:

3 Attachment(s)

Re: Wichman-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

The same for Rnd...
Look carefully there is an image, a line at the top and a line at the bottom. 33554432 is twice the resolution of Rnd and it's distribution is perfect (it returns exact fractions from 0/16777216 to 16777215/16777216, exactly half the data is 0 the other half is 255)

Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

A spectacular firework show!
How did you come up with the test? Can you provide a full coding?
Can you find out exactly the size of the sequence? It was said more than 10^13 but not sure what is exactly. I guess that is
(30269-1) x (30307-1) x (30323-1) = 27,814,431,486,576 (cannot be more than this number).

Updated:
Yes, the cycle length is 27,814,431,486,576 ~ 2.78 x 10^13
This is the first page of Wichmann-Hill article on this PRNG. (Feb-1982, more than 25 years ago!)

Correction:
Due to a spelling mistake on an MSDN page, the correct name of one of the two authors is Wichmann (not Wichman). I apology for this mistake.

Re: Wichman-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

Quote:

Originally Posted by anhn

if not about 5% slower.

My optimization knowledge goes for compiled code, although I know Static are slower, but I'm also a fan of good coding practises (unless I'm going for the uttermost speed one can do with VB6, and I don't see a reason to do that here).

Anyways, without testing I guess my code's other tricks speed it up enough in comparison once compiled. There is a major difference between compiled code and P-code :) I guess I should test, but all my copypasteable benchmarking code is in my broken laptop's HD -> laziness.

In the other hand, there isn't so much to optimize, you'd have to end up with simpler maths, such as taking more advantage of the 31 bits that are available in a Long. Then find two higher prime numbers to use (because the numbers being used a prime numbers) and use those instead of using three. Less math results in more speed, but may drop the randomness level.

More edit:
&H7FFFFFFF / 171 = 12558383,900584795321637426900585
&H7FFFFFFF / 170 = 12632256,747058823529411764705882

Find two prime numbers that are closest below of those two values and you should have something to test :)

Re: Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

Milk has something more to test:

Code:

Public Function RndM(Optional ByVal Number As Long) As Double Static lngX As Long, lngY As Long, lngZ As Long, blnInit As Boolean Dim dblRnd As Double ' if initialized and no input number given If blnInit And Number = 0 Then ' lngX, lngY and lngZ will never be 0 lngX = (170 * lngX) Mod 12632251 lngY = (171 * lngY) Mod 12558383 'lngX = (171 * lngX) Mod 30269 'lngY = (172 * lngY) Mod 30307 'lngZ = (170 * lngZ) Mod 30323 Else ' if no initialization, use Timer, otherwise ensure positive Number If Number = 0 Then Number = Timer * 60 Else Number = Number And &H7FFFFFFF lngX = (Number Mod 12632251) lngY = (Number Mod 12558383) 'lngX = (Number Mod 30269) 'lngY = (Number Mod 30307) 'lngZ = (Number Mod 30323) ' lngX, lngY and lngZ must be bigger than 0 If lngX > 0 Then Else lngX = 170 If lngY > 0 Then Else lngY = 171 'If lngX > 0 Then Else lngX = 171 'If lngY > 0 Then Else lngY = 172 'If lngZ > 0 Then Else lngZ = 170 ' mark initialization state blnInit = True End If ' generate a random number dblRnd = CDbl(lngX) / 12632251# + CDbl(lngY) / 12558383# '+ CDbl(lngZ) / 30323# 'dblRnd = CDbl(lngX) / 30269# + CDbl(lngY) / 30307# + CDbl(lngZ) / 30323# ' return a value between 0 and 1 RndM = dblRnd - Int(dblRnd) End Function

Less math, but will it work? It runs on my assumption that a prime number is the important factor.

Re: Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() functio

hmm, this sounds like it it might be a job for this. I think it will wipeout Rnd, but I think losing the third value might be a great loss in terms of the randomness.... that being kind of relative depending if you are interested in atomic decay or rolling a dice.

I'll give it a test over the WE. What I've done measures the frequency of fractions over a reasonably large resolution. The data I've gathered is not yet fully analysed, and is fairly limited as order of readings is not taken into account.

The old Rnd has 16'777'216 combinations. The Wichmann Hill can give you 27'814'431'486'576. The new two Long algorithm has a possible 158'640'621'019'500 return values.

To put things in perspective, there are 80'658'175'170'943'878'571'660'636'856'403'766'975'289'505'440'883'277'824'000'000'000'000 (68 digits) possible combinations for a deck of cards.

Re: Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

Merry & Milk,
We may have gone too far.
Actually, 171, 172 and 170 have some relations to 30269, 30307 and 30323.
They have been chosen to ensure that for each individual x,y,z can return full cycle.

Re: Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

Are you able to go further into detail on why those are exactly the numbers that work? Also, how could I figure it out without doing a brute-force test? A quick google didn't help me very much and the article you linked to apparently wants me to pay to see it all.

I'm not very good with theoretical math though, this is why I just did what I did.

Re: Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() functio

I don't want to go with too much details on how those numbers worked, as we need to learn much more theory on statistic and randomness. I did one year on statistics 33 years ago and was a top student in this course but there is nothing left in my head now. I don't want to build another algorithm and prove it works, that is too far over my ability now (may be because I am too old, not smart enough and lazy to think).

I have seen many algorithms on PRNG but perhaps Wichmann-Hill's one is quite simple and machine independent, easy to write in any language. (I also found another version of this algorithm for 16-bit machine, but we don't need to discuss it now).

It does not mean we will use whatever Microsoft uses, but there are many articles that complained about the RAND() function in Excel prior to version 2003 so Microsoft decided to use Wichmann-Hill's algorithm to rewrite the RAND() function. The first version of WH RAND() function in Excel 2003 contains a terrible error likes a joke: it can return a negative number, when reported Microsoft had to release a patch to fix it.

Of course there is no perfect PRNG that is why they are called pseudo. WH's algorithm is now more than 25-year-old (Feb-1982), perhaps no one has found any flaw in it, many people have implemented it in many different languages but I haven't seen it in VB (or I could not find it). During the last 2 months I saw at least 2 people in this forum need to have a random function that is better than VB build-in Rnd() function, so I just try to do it in VB with WH's algorithm after I read a KB in MSDN.

As what I understand, all ix, iy and iz are 3 random numbers in a much smaller ranges, however when they are combined as what W&H did they can give a new random number in a much larger range.

As you already knew all 3 numbers 30269, 30307 and 30323 are prime numbers. There are other 3 primitive numbers 171, 172 and 170 that are well chosen so that the subsequence values of ix in formula
ix = (171 * ix) mod 30269
can cover full range from 1 to 30268, (similar for iy & iz).
We should not waste our time to prove that again here, someone else has done that (but if you want you can do it, it's quite easy because the ranges are quite small).

The randomness will be better if we use more numbers instead of only 3 as ix, iy and iz but W&H proved that 3 is adequate or good enough.

Merry, you are really an expert in code optimization as si_the_geek told me.
Milk, you have a great thought with your test that I could not imagine.

There is one thing very interesting in W-H's article, 25 years ago they dream of generating 1000 numbers a second (it may take 880 years to cover full cycle), but now we can generate many million numbers per second with a normal desktop PC. How fast the technology has grew. It's amazing!:)

PS. Sorry, I have tried to find the full WH's article that may contains 3 pages but could not find it, so in the link I said that is the first page of the article.

Re: Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

You should read this: http://www.15seconds.com/issue/051110.htm
After that you may not want to use the pair Rnd() and Randomize() any more in VB, VBA or VB.NET. People may want to use my function (or Merry's version) in their critical projects.

Additional reading:
http://support.microsoft.com/default...b;en-us;120587
http://msdn.microsoft.com/library/de...mrandomize.asp

Re: Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

Also see the Gaussian Random Number generator links in my sig. There is a uniform pseudo-random number generator in there, using a L'Ecuyer generator with a Bays-Durham shuffle.

Re: Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() functio

I started trying out how long it'd take to go through my latter function submission to see if it repeats the first result after going around, but it looks like it'll take about 11 days to do that on my computer. Not going to happen. I've let it count 5 billion results this far and there hasn't been a repetition of the first result. Of course, that doesn't tell us if some other value has repeated at some point...

Edit
Or maybe I miscalculated, now got half a year estimate on recalculation. Anyways, atleast this got as far as 59 000 000 000 results without repeating the first one.

Re: Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

I think a good PRNG is not just based on how long the cycle is (before it repeates), but the randomness is also very important. There exist some tools on the web that can be used to test a PRNG in many areas, but the way Milk used to test it is quite simple and I like it as it also gives graphical presentation as well.

Re: Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() functio

I would not put too much hold on my pretty pictures, good for obvious bias, not good for more subtle problems.

The Die Hard Test is a thing to check out. I believe it's already considered a little dated but it gives a good idea. It tests 32bit integer values (signed or not does not matter) and it requires ~ a 12MB sample of values to test, floating point data won't work. I was using Int((RndX - 0.5) * 4294967296#) to convert to Long.

Incidentally I found a couple of snippets here and there that suggest the Wichman Hill algorithm to have problems at the extremes of the distribution (by modern standards). I've also seen some implementations that use four 'primitive root modulo n' instead of three.

Following is a list of safe primes and the first few primative roots below the primes square root. None of them will overflow.

1660727 1288 1284 1281 1280 1278 1277 1275 1273 1271 1270
1660559 1285 1282 1281 1279 1277 1276 1273 1270 1269 1262
1660199 1287 1286 1279 1278 1274 1272 1269 1267 1266 1265
1660103 1287 1285 1284 1281 1280 1277 1273 1271 1270 1269
1660007 1287 1285 1284 1283 1280 1279 1278 1276 1274 1273
1659683 1288 1287 1285 1283 1281 1280 1279 1276 1275 1274
1659443 1288 1284 1283 1281 1280 1278 1276 1275 1273 1272
1658927 1286 1285 1282 1280 1279 1278 1277 1276 1272 1271
1658387 1287 1285 1284 1283 1280 1276 1275 1274 1272 1271
1658243 1286 1285 1284 1280 1279 1277 1276 1275 1273 1270
1657463 1285 1284 1283 1281 1280 1277 1275 1274 1271 1269
1656899 1286 1285 1284 1283 1281 1279 1277 1275 1274 1272
1656203 1285 1284 1283 1282 1281 1280 1279 1275 1274 1273
1656119 1286 1283 1281 1276 1271 1269 1268 1267 1264 1263
1655939 1282 1281 1277 1276 1275 1274 1272 1268 1267 1265
1655807 1286 1285 1284 1281 1280 1279 1278 1277 1275 1273
1655483 1285 1284 1282 1281 1280 1278 1277 1275 1273 1272
1655279 1283 1282 1281 1276 1275 1274 1271 1270 1262 1259
1655099 1286 1285 1284 1283 1282 1279 1276 1274 1272 1267
1654859 1281 1278 1277 1276 1275 1272 1271 1269 1268 1267
1654739 1286 1283 1282 1278 1276 1275 1273 1270 1267 1266
1654427 1285 1284 1283 1280 1279 1278 1276 1271 1269 1266
1654127 1286 1284 1282 1281 1280 1278 1275 1274 1273 1272
1654019 1286 1285 1284 1282 1279 1276 1274 1269 1267 1265
1653959 1286 1285 1282 1279 1276 1275 1274 1272 1270 1268
1653767 1283 1280 1279 1276 1275 1274 1272 1271 1269 1268
1653623 1283 1281 1280 1279 1272 1271 1270 1268 1267 1266
1653599 1283 1282 1281 1278 1276 1273 1268 1267 1266 1264

Re: Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() functio

I wonder whether this provides solutions to some issues with the last version of RndM (these problems became visible in this thread).

Code:

Public Function RndM(Optional ByVal Number As Long) As Variant Static X As Long, Y As Long, Z As Long, Init As Boolean Dim R As Variant If Init And Number = 0 Then X = (170 * X) Mod 12632251 Y = (171 * Y) Mod 12558383 Z = (172 * Z) Mod 12485359 Else If Number = 0 Then Number = Timer * 60 Else Number = Number And &H7FFFFFFF X = (Number Mod 12632251) Y = (Number Mod 12558383) Z = (Number Mod 12485359) If X = 0 Then X = 170 If Y = 0 Then Y = 171 If Z = 0 Then Z = 172 Init = True End If R = CDec(X) / CDec(12632251) + CDec(Y) / CDec(12558383) + CDec(Z) / CDec(12485359) RndM = R - Int(R) End Function

Instead of speed this is an attempt to provide an even larger scale. Unfortunately my theoretical math skills aren't that great so I'm not going to attempt to figure out how big the maximum set of numbers is.

Re: Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() functio

For people other than Merri: This discussion related to post#23 in this thread
@Merri,
Some things need to be proved first:

Code:

X = (170 * X) Mod 12632251 '-- must cover full range from 1 to 12632250 Y = (171 * Y) Mod 12558383 '-- must cover full range from 1 to 12558382 Z = (172 * Z) Mod 12485359 '-- must cover full range from 1 to 12485358

If you can prove that then the size of the set of numbers returned will be equal to:
12632250 * 12558382 * 12485358 = 1,980,684,946,770,782,481,000
that is still well under 90¹² = 282,429,536,481,000,000,000,000 of the other thread.

So you need to choose 3 other larger prime numbers with a proof as mentioned above.

Secondly, instead of return Scale-28 Decimal numbers (with up to 28 decimal places), we should switch to Scale-0 Decimal numbers (no decimal place) by changing the return formula of R:

Code:

from: R1 = CDec(X) / CDec(12632251) + CDec(Y) / CDec(12558383) + CDec(Z) / CDec(12485359) to: R2 = CDec(x) * CDec(12558383) * CDec(12485359) + _ CDec(Y) * CDec(12485359) * CDec(12632251) + _ CDec(Z) * CDec(12632251) * CDec(12558383) you may see that: R2 = R1 * (CDec(12558383) * CDec(12485359) * CDec(12632251))

For not confusing, rename the function to something like LargeRndInt()

Re: Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() functio

I made some searching and found out people prefer Mersenne twister these days. There is a VB6 implementation over at Planet Source Code, the problem being it is not a short code and it isn't optimized in any way. Doesn't interest me enough to go ahead and work around the issues... I guess the unsigned 32-bit processing should be dropped to 31-bit so that it works with Long – but I don't know how much this would change the algorithm. Would require reading & studying to get it right.

Update!
There is also an implementation in REALbasic. Being a relatively short code it could be used as a model for VB6 version.

Re: Wichmann-Hill Pseudo Random Number Generator: an alternative for VB Rnd() functio

Excuse Me! Why More and More Lines of Codes?

Just use this to Generate Random Numbers

Code:

Public Function GenerateCode(iLen As Integer) As String On Error Resume Next strInputString = "0123456789" intLength = Len(strInputString) intNameLength = iLen Randomize strName = "" For intStep = 1 To intNameLength intRnd = Int((intLength * Rnd) + 1) strName = strName & Mid(strInputString, intRnd, 1) Next GenerateCode = strName End Function

Re: Wichman-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

Quote:

Originally Posted by Merri

Here is a suggestion for independent function. It doesn't work exactly the same, but I paid a lot attention to details on how it works.

Code:

Public Function RndM(Optional ByVal Number As Long) As Double Static lngX As Long, lngY As Long, lngZ As Long, blnInit As Boolean Dim dblRnd As Double ' if initialized and no input number given If blnInit And Number = 0 Then ' lngX, lngY and lngZ will never be 0 lngX = (171 * lngX) Mod 30269 lngY = (172 * lngY) Mod 30307 lngZ = (170 * lngZ) Mod 30323 Else ' if no initialization, use Timer, otherwise ensure positive Number If Number = 0 Then Number = Timer * 60 Else Number = Number And &H7FFFFFFF lngX = (Number Mod 30269) lngY = (Number Mod 30307) lngZ = (Number Mod 30323) ' lngX, lngY and lngZ must be bigger than 0 If lngX > 0 Then Else lngX = 171 If lngY > 0 Then Else lngY = 172 If lngZ > 0 Then Else lngZ = 170 ' mark initialization state blnInit = True End If ' generate a random number dblRnd = CDbl(lngX) / 30269# + CDbl(lngY) / 30307# + CDbl(lngZ) / 30323# ' return a value between 0 and 1 RndM = dblRnd - Int(dblRnd) End Function

Here are a few pointers:
True conditions are faster than False conditions -> set code that is ran more often behind True condition.
I minimized the amount of calculation when checking for initialized values. By using a local static variable we know that X, Y and Z are valid.
I decided to use 0 as number to cause Timer to be used automatically for seeding: if someone wants repeatable random numbers, he knows to give a seed.
... which leads to this function having an advantage over Rnd as you don't need to call Randomize to get a different set of random numbers on each run of the application.
And I used variable names that tell better what datatype they represent.

Hi Merri, Not sure of the code rules here but is it ok if I use this code in my project?

Re: Wichman-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

Quote:

Originally Posted by fruitman

Hi Merri, Not sure of the code rules here but is it ok if I use this code in my project?

I think it is fine to use other people code in your project as long as you include credit to the appropriate people/group of in your application.

Re: Wichman-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

Dear Mr. Piittinen,

I am using your algorithm "rndM" (Wichmann-Hall random number generator) [1] for educational purposes and would like to include it in a book, if you don't object.

I have written a book (in German): "Physics with Excel and visual basic" [2] and I am preparing a second book (also in German) "Learning and understanding Physics with Excel and visual basic", where I treat topics in Theoretical Physics taught in the first two years of a Physics course at universities. In the chapter "Stochastic movements" I realized the limits of the VBA function RND. The entropy in a simulated diffusion process began to decrease after some diffusion time while your algorithm yielded reasonable results.

In the first book, I discovered also an error in the LOGEST spreadsheet function. The uncertainties of the regression coefficients were wrong but could be corrected with simple operations. I found nowhere in Microsoft to report this.

With kind regards
Dieter Mergel

[1] http://www.vbforums.com/showthread.p...Rnd()-function, by user Merri Dec 6th, 2007, 12:29 AM (Vesa Piittinen)
[2] http://www.springer.com/de/book/9783642378560

Re: Wichman-Hill Pseudo Random Number Generator: an alternative for VB Rnd() function

Dear Mr. Piittinen,

I am using your algorithm "rndM" (Wichmann-Hall random number generator) [1] for educational purposes and would like to include it in a book, if you don't object.

I have written a book (in German): "Physics with Excel and visual basic" [2] and I am preparing a second book (also in German) "Learning and understanding Physics with Excel and visual basic", where I treat topics in Theoretical Physics taught in the first two years of a Physics course at universities. In the chapter "Stochastic movements" I realized the limits of the vba function RND. The entropy in a simulated diffusion process started to decrease after some diffusion time while your algorithm yielded reasonable results.

In the first book, I discovered also an error in the LOGEST spreadsheet function. The uncertainties of the regression coefficients were wrong but could be corrected with simple operations. I found nowhere in Microsoft to report this.

With kind regards
Dieter Mergel

[1] http://www.vbforums.com/showthread.p...Rnd()-function, by user Merri Dec 6th, 2007, 12:29 AM (Vesa Piittinen)
[2] http://www.springer.com/de/book/9783642378560

Quote:

Originally Posted by Merri

Here is a suggestion for independent function. It doesn't work exactly the same, but I paid a lot attention to details on how it works.

Code:

Public Function RndM(Optional ByVal Number As Long) As Double Static lngX As Long, lngY As Long, lngZ As Long, blnInit As Boolean Dim dblRnd As Double ' if initialized and no input number given If blnInit And Number = 0 Then ' lngX, lngY and lngZ will never be 0 lngX = (171 * lngX) Mod 30269 lngY = (172 * lngY) Mod 30307 lngZ = (170 * lngZ) Mod 30323 Else ' if no initialization, use Timer, otherwise ensure positive Number If Number = 0 Then Number = Timer * 60 Else Number = Number And &H7FFFFFFF lngX = (Number Mod 30269) lngY = (Number Mod 30307) lngZ = (Number Mod 30323) ' lngX, lngY and lngZ must be bigger than 0 If lngX > 0 Then Else lngX = 171 If lngY > 0 Then Else lngY = 172 If lngZ > 0 Then Else lngZ = 170 ' mark initialization state blnInit = True End If ' generate a random number dblRnd = CDbl(lngX) / 30269# + CDbl(lngY) / 30307# + CDbl(lngZ) / 30323# ' return a value between 0 and 1 RndM = dblRnd - Int(dblRnd) End Function

Here are a few pointers:
True conditions are faster than False conditions -> set code that is ran more often behind True condition.
I minimized the amount of calculation when checking for initialized values. By using a local static variable we know that X, Y and Z are valid.
I decided to use 0 as number to cause Timer to be used automatically for seeding: if someone wants repeatable random numbers, he knows to give a seed.
... which leads to this function having an advantage over Rnd as you don't need to call Randomize to get a different set of random numbers on each run of the application.
And I used variable names that tell better what datatype they represent.