[Random] uniform_01

[Robert Patterson-2]

Hi,

I was looking at the docs for the uniform_01 distribution:

http://www.boost.org/doc/libs/1_39_0/libs/random/random-distributions.html#uniform_01

I read the following note
"Note: The current implementation is buggy, because it may not fill all
of the mantissa with random bits."

I do not understand the problem.  I looked at the code, which does what
I expect.  Why is it a problem if some of the mantissa bits are not
'random'?  As I understand the requirement is the real numbers are
uniformly distributed; the bit patterns should not matter provided they
represent numbers with the right mathematical properties.

Thanks

Robert Patterson

http://como.cheng.cam.ac.uk/index.php?Page=People&Section=riap2


_______________________________________________
Boost-users mailing list
Boost-users@...
http://lists.boost.org/mailman/listinfo.cgi/boost-users

[Oei, YC]

Hi Rob!

> I do not understand the problem.  I looked at the code, which does what
> I expect.  Why is it a problem if some of the mantissa bits are not
> 'random'?  As I understand the requirement is the real numbers are
> uniformly distributed; the bit patterns should not matter provided they
> represent numbers with the right mathematical properties.

As far as I understand (sorry for starting with a big disclaimer but
I'm mostly just a user), the thing is that uniform_01 does not change
its behaviour if you couple it with a generator that produces 32bit
integers (like the mersenne twister ones). If you need sufficient
random bits to fill a double with, you should really be drawing two
integers from the generator, but I don't think that's what happens.

Yung-Chin
_______________________________________________
Boost-users mailing list
Boost-users@...
http://lists.boost.org/mailman/listinfo.cgi/boost-users

[Anton Daneyko]

Indeed this vague mention leaves you with no idea of weather you can use that or not. It would be great if this note would be made more verbose.

On Thu, Jul 9, 2009 at 1:59 PM, Robert Patterson <riap2@...> wrote:

Hi,

I was looking at the docs for the uniform_01 distribution:

http://www.boost.org/doc/libs/1_39_0/libs/random/random-distributions.html#uniform_01

I read the following note
"Note: The current implementation is buggy, because it may not fill all
of the mantissa with random bits."

I do not understand the problem.  I looked at the code, which does what
I expect.  Why is it a problem if some of the mantissa bits are not
'random'?  As I understand the requirement is the real numbers are
uniformly distributed; the bit patterns should not matter provided they
represent numbers with the right mathematical properties.

Thanks

Robert Patterson

http://como.cheng.cam.ac.uk/index.php?Page=People&Section=riap2


_______________________________________________
Boost-users mailing list
Boost-users@...
http://lists.boost.org/mailman/listinfo.cgi/boost-users

_______________________________________________
Boost-users mailing list
Boost-users@...
http://lists.boost.org/mailman/listinfo.cgi/boost-users