Random number in Fortran 90/95
in [Fortran]
|
From: Uno on 27 Jun 2010 20:19 Richard Maine wrote: > Uno <merrilljensen(a)q.com> wrote: > [about the vagueness in Fortran's definition of random numbers] >> It sounds like an opportunity squandered. I'd be curious what you would >> like it to have said. > > I have not bothered to try to craft appropriate words that aren't going > to get used. Nor do I plan to. It seems like a pointless exercise. For > similar reasons, I haven't even wasted much time in thinking about how I > would have liked Fortran's random number stuff to have worked. I don't > even recall for sure exactly what the author of the words in the Fortran > standard said that he intended. There are at least some arguments on > multiple sides (at least 2 - maybe more). > > The only thing I have to say that is that I would have liked for > whatever it said to have been unambiguous. Is there something ambiguous about the absence of words? > >> In C, random numbers are ints between 0 and RAND_MAX. Keith Thompson >> writes that it simply underlaid the process that these numbers were to >> be random. It's not written in their standard anywhere. > > Keith is normally more precise than that. I suppose that he might have > been sloppy enough to refer to numbers being "random", but that would at > least minorly surprise me. I'd guess it more likely that you are > misquoting him. Note that "pseudorandom" is not the same thing as > "random." No, he's mathematically as helpless as you are. What he said was in context, but I think I got the gist, and I did consider being an appropriate editor when I typed it. > >> Do pseudorandoms in fortran have to have a flat pdf on [0,1) according >> to the fortran standard? > > Is there some reason you don't just look in the standard or one of the > references about it instead of asking me? It isn't as though this > particular information isn't in other that the most obvious place or is > stated in other than the most obvious manner. > > I often try to give answers to questions of subtle (or even > not-so-subtle) interpretation or of just finding something that is > buried in an obscure place in the standard. Those kinds of questions > take advantage of my experience/expertise/whatever. But I don't see why > I should provide straightforward reading service. If I wanted to make > sure that my answer was actually correct, I'd have to go look it up, > just like you would. On doing so, I do see that my recollection was > correct, but doing the same check as I did is left as an exercise for > the student on the principle of teaching him how to fish. > http://i45.tinypic.com/23v0xsh.png I find none of this straightforward reading. Whatever comes out of the can to read a .pdf on ubuntu has a *terrible* search capability. I've tried to upgrade this capacity but have failed twice. -- Uno
From: Jason Riedy on 28 Jun 2010 18:04 And Richard Maine writes: > I often try to give answers to questions of subtle (or even > not-so-subtle) interpretation or of just finding something that is > buried in an obscure place in the standard. Here's one for you, which I cannot find but expect is not specified: Do you happen to know if the statistical interplay between random numbers and images are specified? I haven't found anything, and I suspect the issue was either ignored or so contentious that it wasn't included. Considering the spec for random_number doesn't mention images or co-arrays, users could assume a fancy parallel generator. (Or my UPC background may color my expectations for passing co-arrays to subroutines.) So many people take parallel random number generation for granted... Jason
From: Richard Maine on 28 Jun 2010 18:21 Jason Riedy <jason(a)acm.org> wrote: > And Richard Maine writes: > > I often try to give answers to questions of subtle (or even > > not-so-subtle) interpretation or of just finding something that is > > buried in an obscure place in the standard. > > Here's one for you, which I cannot find but expect is not specified: Do > you happen to know if the statistical interplay between random numbers > and images are specified? I haven't found anything, and I suspect the > issue was either ignored or so contentious that it wasn't included. I haven't been directly involved with much after f2003, and in particular, with the coarray stuff. (In fact, until I saw your later mention of coarrays, I didn't pick up what you were talking about when you said "images". Nothing wrong with your usage; I just haven't been imersed in it enough that the term makes coarray stuff imediately pop into my mind. My initial reaction was to wonder why anyone thought the Fortran standard would say anything about picture images.) My off-hand guess is that nobody even thought about it, but that's not based on any real data. It is just the kind of thing I'd have expected to get overlooked. -- Richard Maine | Good judgment comes from experience; email: last name at domain . net | experience comes from bad judgment. domain: summertriangle | -- Mark Twain
From: glen herrmannsfeldt on 28 Jun 2010 18:28 Jason Riedy <jason(a)acm.org> wrote: (snip) > Here's one for you, which I cannot find but expect is not specified: Do > you happen to know if the statistical interplay between random numbers > and images are specified? I haven't found anything, and I suspect the > issue was either ignored or so contentious that it wasn't included. (from Fortran 2003) "Returns one pseudorandom number or an array of pseudorandom numbers from the uniform distribution over the range 0 x < 1." (There is a <= in there that may or may not show up.) It doesn't say anything about the quality if the generated numbers, other than the range and uniform distribution. > Considering the spec for random_number doesn't mention images or > co-arrays, users could assume a fancy parallel generator. (Or my UPC > background may color my expectations for passing co-arrays to > subroutines.) So many people take parallel random number generation for > granted... I believe co-arrays came in Fortran 2008, so you might look there. It would seem that two possibilities are a separate generator (seed) for each image, or a synchronization process for a single generator. With appropriate seeding the separate generators should generate a repeatable sequence, where the synchronized single generator likely won't. My guess is that this is a quality of implementation issue. (or user preference, for that matter) -- glen
From: GaryScott on 28 Jun 2010 18:43
On Jun 28, 5:21 pm, nos...(a)see.signature (Richard Maine) wrote: > Jason Riedy <ja...(a)acm.org> wrote: > > And Richard Maine writes: > > > I often try to give answers to questions of subtle (or even > > > not-so-subtle) interpretation or of just finding something that is > > > buried in an obscure place in the standard. > > > Here's one for you, which I cannot find but expect is not specified: Do > > you happen to know if the statistical interplay between random numbers > > and images are specified? I haven't found anything, and I suspect the > > issue was either ignored or so contentious that it wasn't included. > > I haven't been directly involved with much after f2003, and in > particular, with the coarray stuff. (In fact, until I saw your later > mention of coarrays, I didn't pick up what you were talking about when > you said "images". snip So many times, the terminology choices are unfortunate. Maybe it is the lack of depth of experience of the very small sized committee. I don't think you choose such terminology that for the vast majority has the meaning of a "bitmapped" picture image. Maybe a modifier would be better (execution image?), but then the modifier will get lost as people get sloppy. > -- > Richard Maine | Good judgment comes from experience; > email: last name at domain . net | experience comes from bad judgment. > domain: summertriangle | -- Mark Twain |