www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Random distributions in Phobos

reply Joseph Rushton Wakeling <joseph.wakeling webdrake.net> writes:
Hello all,

Currently Phobos only implements the uniform distribution for random number 
generation.  It should be relatively trivial to extend that to other 
distributions (exponential, normal, Pareto, ...), so I'm considering putting 
together some code.

Before I start writing, though: are there any standards or specifications that 
are being followed in constructing these kinds of mathematical functions for
Phobos?

Thanks & best wishes,

     -- Joe
Apr 24 2012
parent reply "Jesse Phillips" <jessekphillips+D gmail.com> writes:
On Wednesday, 25 April 2012 at 01:03:39 UTC, Joseph Rushton 
Wakeling wrote:

 Hello all,

 Currently Phobos only implements the uniform distribution for 
 random number generation.  It should be relatively trivial to 
 extend that to other distributions (exponential, normal, 
 Pareto, ...), so I'm considering putting together some code.

 Before I start writing, though: are there any standards or 
 specifications that are being followed in constructing these 
 kinds of mathematical functions for Phobos?

 Thanks & best wishes,

     -- Joe


I suppose it would make sense for these to make it into phobos, 
personally am not familiar with the use case.

Instead of writing the answer to your question here, I've made 
changes to the wiki. I think there is a page I'm missing but 
don't know where it is so maybe someone else will correct it:

http://www.prowiki.org/wiki4d/wiki.cgi?HelpDProgress#ContributingtoPhobos
Apr 25 2012
next sibling parent reply Joseph Rushton Wakeling <joseph.wakeling webdrake.net> writes:
On 26/04/12 06:03, Jesse Phillips wrote:

 I suppose it would make sense for these to make it into phobos, personally am
 not familiar with the use case.


The use case is mostly to do with scientific simulation: if you look at most 
science-oriented languages and libraries (MATLAB/Octave, R, GNU Scientific 
Library, ...) they offer an extensive range of different random number 
distributions.

SciD would also be an acceptable location for this kind of functionality, but 
going by the example of e.g. Boost.Random it seems appropriate to have the
basic 
RNG functionality coupled with extra distributions.  It's also clear from the 
std.random documentations that more distributions are planned for inclusion.


 Instead of writing the answer to your question here, I've made changes to the
 wiki. I think there is a page I'm missing but don't know where it is so maybe
 someone else will correct it:

 http://www.prowiki.org/wiki4d/wiki.cgi?HelpDProgress#ContributingtoPhobos


OK, thanks.  So to be clear: I should submit my proposed changes as a pull 
request, making sure to include a and should expect feedback after about 2
weeks 
... ?

I ask because I wasn't clear if I'd done the right thing when I submitted a
pull 
request on my random sampling functionality.  I was expecting to get at least
an 
acknowledgement quite quickly, either saying that the code would be looked at
or 
highlighting an obvious missing factor (e.g. appropriate unittests or
benchmarks).

Thanks & best wishes,

     -- Joe
Apr 26 2012
parent "Jesse Phillips" <Jessekphillips+D gmail.com> writes:
On Thursday, 26 April 2012 at 11:46:32 UTC, Joseph Rushton 
Wakeling wrote:

 OK, thanks.  So to be clear: I should submit my proposed 
 changes as a pull request, making sure to include a and should 
 expect feedback after about 2 weeks ... ?


Good follow up, I actually forgot about the three types of 
submissions and yours being the second (bugs third).

The review processes is really for major additions or changes. 
You are suggesting an expansion on existing functionality.

In which case submission via a pull request before review is 
acceptable, the review is done by the Phobos maintainers instead 
of requiring whole community input. However once you have a pull 
request ready, I think it is good to post to the forum to give it 
more visibility.

I also think adding a bugzilla entry for the enhancement may also 
be welcomed (other opinions?). The pull request would then state 
it closes the bug.

http://d.puremagic.com/issues/

As for response expectations, there can't really be. There is a 
good chance things will sit in the queue for some time, if it 
takes too long, then asking again I don't see as being frowned 
upon.
Apr 26 2012
prev sibling parent "Jonathan M Davis" <jmdavisProg gmx.com> writes:
On Thursday, April 26, 2012 13:46:20 Joseph Rushton Wakeling wrote:

 On 26/04/12 06:03, Jesse Phillips wrote:

 I suppose it would make sense for these to make it into phobos, personally
 am not familiar with the use case.


 
 The use case is mostly to do with scientific simulation: if you look at most
 science-oriented languages and libraries (MATLAB/Octave, R, GNU Scientific
 Library, ...) they offer an extensive range of different random number
 distributions.
 
 SciD would also be an acceptable location for this kind of functionality,
 but going by the example of e.g. Boost.Random it seems appropriate to have
 the basic RNG functionality coupled with extra distributions. It's also
 clear from the std.random documentations that more distributions are
 planned for inclusion.

 Instead of writing the answer to your question here, I've made changes to
 the wiki. I think there is a page I'm missing but don't know where it is
 so maybe someone else will correct it:
 
 http://www.prowiki.org/wiki4d/wiki.cgi?HelpDProgress#ContributingtoPhobos


 
 OK, thanks. So to be clear: I should submit my proposed changes as a pull
 request, making sure to include a and should expect feedback after about 2
 weeks ... ?
 
 I ask because I wasn't clear if I'd done the right thing when I submitted a
 pull request on my random sampling functionality. I was expecting to get
 at least an acknowledgement quite quickly, either saying that the code
 would be looked at or highlighting an obvious missing factor (e.g.
 appropriate unittests or benchmarks).


No. The whole "2 weeks" thing is talking about the review process for adding 
major new functionality to Phobos (e.g. adding a new module). Major stuff needs 
to be reviewed and voted in on the newsgroup per the Boost review process (or 
something approximating it anyway). After something has been reviewed and 
voted it, then it goes through the normal pull request process to actually get 
merged into Phobos.

Smaller stuff (e.g. fixing a bug or adding one function) can go through the
pull 
request process without a review in the newsgroup (the Phobos devs will 
normally point it out if something is major enough to need full review if 
you're not sure).

However, there is no guarantee whatsoever about how quickly a pull request 
will be processed. Sometimes, it's very quick. Other times, a pull request can 
sit there for a while before it gets looked at. The Phobos devs are all 
volunteers working in their own time, and there are only so many of them, so 
when they get to pull requests tends to be very dependent on their personal 
schedules and on what the pull request is for (e.g. if it's for something that 
a particular developer works on regularly or it's very simple, it's much more 
likely to be processed quickly, but if it's more esoteric and/or large and 
complicated, then it's much more likely to take a while).

- Jonathan M Davis
Apr 26 2012