Is there a way to generate Gaussian random numbers in Ocean/Skill?

Hi,

i think you can create using random() a (nearly ?) uniform distributed random variable. By using two uniform distributed random variables and Box-Muller transform you can create two normal distributed variables. By scaling with sigma and mu you can get your desired normal distribution. Please note that I haven't checked it myself.

Kind regards,

Matthias

Yes - agreed that you can always use existing box-car random numbers and, if you write your own function, make something that yields a a normal distribution. However, it's such a fundamental thing you'd reckon the function would exist somewhere within the Skill/Ocean world. You could also write your own code to do logarithms and trigonometric functions, but happily they exist already (so OK, maybe that comment was a little harsh)...

I wouldn't class this the same way as log or trig functions, since those are rather more fundamental (for example, those are in the C maths library, but a Gaussian random number generator isn't AFAIK). However, I have written a function to do this a while back (using the Box-Muller method) - which can be found in this post.

Regards,

Andrew.

I actually did find your code by searching further and am already using it successfully - so yet another thank you for your efforts! I did mean to respond to this post after I found it but forgot. I also took a look at your code (presumably written in Lisp or something) and couldn't understand a bloody word of it.

Captain's pesronal log:  Bummer that Cadence has to use such an ancient, miserable, obtuse language like that. I think that's how their applications engineers get 100% guaranteed job security ;-).

CADcasualty said:

Bummer that Cadence has to use such an ancient, miserable, obtuse language like that

Yes, LISP is an old language, but it's not necessarily obtuse (and I'm not being overly defensive here). I tend to write my code in LISP style because when I first started writing SKILL (before I worked for Cadence), I had written a lot of C and the fact that the C-style wasn't quite C tended to cause me to make mistakes, plus it encouraged me to think using LISP approaches to problems. I'd also written some LISP at University too...

Using the pp() function you can easily dump out the code in a more familiar C style. In fact using this code you can convert from LISP to C style:

procedure(abConvToC(inFile outFile)
  let((inPort outPort (data t))
    inPort=infile(inFile)
    outPort=outfile(outFile)
    while(data
      while((data=lineread(inPort))==t t)
      foreach(form data
        pprint(form outPort)
        newline(outPort)
      )
    )
    close(inPort)
    close(outPort)
  )
)

So that means the code (I didn't bother with the test function) above is:

defMathConstants('abRandomNormal)
defun(abRandomNormal nil 
    let((max((-1 >> 1)) U V) 
	(U = (float(random(max)) / max)) 
	(V = (float(random(max)) / max)) 
	(sqrt((-2 * log(U))) * cos((2 * (abRandomNormal.PI) * V)))
    )
)

The pretty printer puts a few extra brackets than strictly necessary, but hopefully it's a bit easier to understand. Probably the only complicated bit still is that max is initialised in the let to be -1 >> 1 which is the maximum integer (I could have used abRandomNormal.INT_MAX instead - defMathConstants defines that usefully as well as defining PI).

I completely get that people don't necessarily find the LISP format easy to understand, which is why I write code intended as examples (i.e. something you're going to modify or base your own code upon) in C-style. However, for code that is likely to be used as-is as a library function, I write it the way I find more natural. I would say that most AEs in Cadence don't write using LISP style though.

Regards,

Andrew.