Re: RND contest


Betov wrote:
"Alex McDonald" <alex_mcd@xxxxxxxxxxxxxxx> écrivait
news:1140522662.452544.202110@xxxxxxxxxxxxxxxxxxxxxxxxxxxx:

< http://betov.free.fr/RndCompare.exe >

... shows... nice vertical lines... that my eyes
can very well see.

Have you tried changing the size of the window? As if by magic, the
lines disappear. In fact, I've saved a screenshot for you;
http://www.rivadpm.com/betov/random.bmp. To my eyes, it /looks/ random.


The point is that this is not a random series; quite the opposite. It's
the series 7*11, 7*11*11, 7*11*11*11 etc.

Congratuations: You have found a way for perverting
the test, so that a bad Algo could look like better,
occasionaly.

Now, the problem has never been to find out a way
to obfuscate the results. The fact is that, with
some sizes of the Window your Routine shows vertical
lines, whereas the other ones never do.

Then, with most Sizes, your Routine still shows lines.
Of various kinds, smaller of bigger, like this or like
that, and even lines that are very difficult to see,
but still _lines_, that show clearly the deffect.

You have demonstrated that the test is probably not the
best possible one, and/or that various sizes should be
given a try for each Routine.

And now? Are you going to extend the Random Concept to
"Looking Random under some particular condition", or
to consider that one single condition showing the deffect
is enough for demonstrating the deffect?

:))

Can you find out some size of the window that would show
Lines, for the two other Routines?

:))


That wasn't my point. I'm not claiming your routine isn't generating
pseudo-random numbers. What I said was that visual tests of randomness
aren't useful; the eye misinterprets the data. You're testing for, and
seeing, uniform distribution. That's quite different from randomness,
as the example shows.

--
Regards
Alex McDonald

.

Relevant Pages

  • Re: RND contest
    ... Have you tried changing the size of the window? ... Then, with most Sizes, your Routine still shows lines. ... that show clearly the deffect. ...
    (alt.lang.asm)
  • Re: OE has gone large print
    ... So, tomorrow, I'll once again try to copy all my .dbx folders from old 'defective' identity to this new one. ... All of a sudden, the scroll bar wouldn't move, the entire window began to shimmy. ... the font size is out of whack - too big. ... Some old e-mails look ok, but if I attempt to forward and add content, my normal Arial 12 looks two sizes bigger. ...
    (microsoft.public.windows.inetexplorer.ie6_outlookexpress)
  • Re: Windows Media Tools Corruption
    ... Tried the routine you suggested In ZachD's site - to no avail. ... weird if it also does that in WMP9 menus too, ... in the command window type ...
    (microsoft.public.windowsmedia)
  • Re: VB for Office program flow question
    ... > Does the closing of the new form return control to the routine that> opened it? ... > open window and wait for user input ...
    (microsoft.public.access.formscoding)
  • OE has gone large print
    ... inbox mail in the view window. ... the entire window began to shimmy. ... the font size is ... my normal Arial 12 looks two sizes bigger. ...
    (microsoft.public.windows.inetexplorer.ie6_outlookexpress)
Loading