Re: Re: Mathematica 7 weirdness
- To: mathgroup at smc.vnet.net
- Subject: [mg94607] Re: [mg94566] Re: Mathematica 7 weirdness
- From: Andrzej Kozlowski <akoz at mimuw.edu.pl>
- Date: Wed, 17 Dec 2008 06:35:09 -0500 (EST)
- References: <C7EFEF88-F3D3-4E38-87D4-E481FFF62029@mimuw.edu.pl> <200812160736.CAA13469@smc.vnet.net> <4947D8E0.7020309@wolfram.com>
On 17 Dec 2008, at 01:35, Darren Glosemeyer wrote:
> Andrzej Kozlowski wrote:
>> On 16 Dec 2008, at 15:42, Andrzej Kozlowski wrote:
>>
>>
>>> Here is a (slightly modified) example from the
>>> RandomNumberGeneration tutorial, which comes as part of the
>>> documentation:
>>>
>>> BlockRandom[
>>> SeedRandom[Method -> {"MKL", Method -> {"Niederreiter",
>>> "Dimension" -
>>>> 2}}];
>>>>
>>> RandomReal[1, {2, 2}]]
>>>
>>> The problem is that as well as the expected output we get an
>>> error message:
>>>
>>> SeedRandom::nogen:MKL is not one of the built-in random number
>>> generators Lattice, Congruential, MersenneTwister, Rule30CA,
>>> Rule50025CA, ExtendedCA, or Legacy. A defined generator must be
>>> represented by a symbol. >>
>>>
>>> One possible reason is that I am using Mac OS X and the
>>> documentation says:
>>>
>>> The "MKL" method uses the random number generators provided in
>>> Intel's MKL libraries. The MKL libraries are platform dependent.
>>> The "MKL" method is available on Microsoft Windows (32-bit, 64-
>>> bit), Linux x86 (32-bit, 64-bit), and Linux Itanium systems.
>>>
>>> Intel Macs are not mentioned, but since, other than producing the
>>> error message, the generator works fine, this should count as a
>>> serious omission (if not exactly a bug) which should be corrected
>>> as soon as possible.
>>>
>>>
>>> Andrzej Kozlowski
>>>
>>
>> On second thoughts, since all that is needed is
>>
>> Off[SeedRandom::"nogen"]
>>
>> probably I should have written "not very serious omission" ;-)
>>
>> Andrzej Kozlowski
>>
>>
>
> I checked with a colleague who uses an Intel Mac and he verified
> that the number generator is falling back to the default method.
> While numbers are generated, the numbers generated will not have the
> low discrepancy properties of Niederreiter numbers.
>
> Darren Glosemeyer
> Wolfram Research
Yes, one can actually clearly see that this must be so just by looking
at the picture :-(
This is very disappointing. I was planning to low discrepancy number
in a series of presentations which aim at popularizing Mathematica
with the Japanese finance community, but it now seems I will have to
settle for Monte-Carlo methods only (since I do not want to use
Windows or Linux). Is this something that could be easily (and
quickly) fixed? In particular, can it be done by WRI alone or does it
need help from Apple (if it is the latter I doubt the fix will come
soon).
Andrzej Kozlowski
- References:
- Re: Mathematica 7 weirdness
- From: Andrzej Kozlowski <akoz@mimuw.edu.pl>
- Re: Mathematica 7 weirdness