[random] Speed up binomial and poisson distributions
Hi, I was wondering why the binomial distribution still isn't optimized for speed. If I look into the GSL as a comparison, they use much smarter and faster algorithms. As is, the bosst::random distributions are not usable for computational purposes to me, which is a pity since I very much like the approach using variate_generators and stuff... I would be happy to contribute some faster code, but I don't know how exactly I should proceed for that... should I contact the maintainer directly? Stephan
-----Original Message----- From: boost-bounces@lists.boost.org [mailto:boost-bounces@lists.boost.org] On Behalf Of Stephan Sent: 06 March 2009 10:26 To: boost@lists.boost.org Subject: [boost] [random] Speed up binomial and poisson distributions
Hi,
I was wondering why the binomial distribution still isn't optimized for speed. If I look into the GSL as a comparison, they use much smarter and faster algorithms. As is, the bosst::random distributions are not usable for computational purposes to me, which is a pity since I very much like the approach using variate_generators and stuff...
I would be happy to contribute some faster code, but I don't know how exactly I should proceed for that... should I contact the maintainer directly?
Yes - I think you should - and report back here if you fail - someone else may be willing to take on the task - or help *you* to take it on ;-) Paul --- Paul A. Bristow Prizet Farmhouse Kendal, UK LA8 8AB +44 1539 561830, mobile +44 7714330204 pbristow@hetp.u-net.com
Hey Stephan, I have a quick poisson generator based on Ahrens and Dieter's PD algorithm. I've been too busy to formally submit it. If you want to take a look it's here: http://dl.getdropbox.com/u/416909/poisson_distribution.hpp If anyone knows how to submit this for review, please let me know. Ross Levine On Sat, Mar 7, 2009 at 5:04 AM, Paul A. Bristow <pbristow@hetp.u-net.com> wrote:
-----Original Message----- From: boost-bounces@lists.boost.org [mailto:boost-bounces@lists.boost.org] On Behalf Of Stephan Sent: 06 March 2009 10:26 To: boost@lists.boost.org Subject: [boost] [random] Speed up binomial and poisson distributions
Hi,
I was wondering why the binomial distribution still isn't optimized for speed. If I look into the GSL as a comparison, they use much smarter and faster algorithms. As is, the bosst::random distributions are not usable for computational purposes to me, which is a pity since I very much like the approach using variate_generators and stuff...
I would be happy to contribute some faster code, but I don't know how exactly I should proceed for that... should I contact the maintainer directly?
Yes - I think you should - and report back here if you fail - someone else may be willing to take on the task
- or help *you* to take it on ;-)
Paul
--- Paul A. Bristow Prizet Farmhouse Kendal, UK LA8 8AB +44 1539 561830, mobile +44 7714330204 pbristow@hetp.u-net.com
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
First of all, have you tried to contact the original author Jens Maurer http://www.boost.org/users/people/jens_maurer.html gives an email address. Or does anyone else have a current address? Paul
-----Original Message----- From: boost-bounces@lists.boost.org [mailto:boost-bounces@lists.boost.org] On Behalf Of Ross Levine Sent: 09 March 2009 05:46 To: boost@lists.boost.org Subject: Re: [boost] [random] Speed up binomial and poisson distributions
Hey Stephan,
I have a quick poisson generator based on Ahrens and Dieter's PD algorithm. I've been too busy to formally submit it. If you want to take a look it's here: http://dl.getdropbox.com/u/416909/poisson_distribution.hpp
If anyone knows how to submit this for review, please let me know.
Ross Levine
On Sat, Mar 7, 2009 at 5:04 AM, Paul A. Bristow <pbristow@hetp.u-net.com> wrote:
-----Original Message----- From: boost-bounces@lists.boost.org
[mailto:boost-bounces@lists.boost.org]
On
Behalf Of Stephan Sent: 06 March 2009 10:26 To: boost@lists.boost.org Subject: [boost] [random] Speed up binomial and poisson distributions
Hi,
I was wondering why the binomial distribution still isn't optimized for speed. If I look into the GSL as a comparison, they use much smarter and faster algorithms. As is, the bosst::random distributions are not usable for computational purposes to me, which is a pity since I very much like the approach using variate_generators and stuff...
I would be happy to contribute some faster code, but I don't know how exactly I should proceed for that... should I contact the maintainer directly?
Yes - I think you should - and report back here if you fail - someone else may be willing to take on the task
- or help *you* to take it on ;-)
Paul
--- Paul A. Bristow Prizet Farmhouse Kendal, UK LA8 8AB +44 1539 561830, mobile +44 7714330204 pbristow@hetp.u-net.com
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
I sent Jens an email just now. I have a style questions as well. First of all, this new poisson_distribution is policy-based (don't worry -- it's backwards-compatible with the old one; old code using the poisson_distribution will not need to be changed). Since poisson_distribution resides in the boost namespace, I didn't know if the policies should be placed there as well, or in the boost/random nested namespace. For now, they are in the boost/random/detail namespace and typedef'ed into the boost namespace. They are named pd_fast, pd_small, and pd_legacy. However, it occurred to me that perhaps having policy classes in the boost namespace wasn't such a good idea, since these classes are not meant to be instatiated directly. At the very least, they could have private constructors and be friends with the poisson_distribution, but I still don't know if they ought to reside in the main boost namespace. If anyone knows any rules for this, let me know. As it is there isn't much mention of this situation in the developer faq. In addition, this header probably needs formal documentation. I am unaware of any stylistic guidelines I should follow. Thanks, Ross On Mon, Mar 9, 2009 at 7:49 AM, Paul A. Bristow <pbristow@hetp.u-net.com> wrote:
First of all, have you tried to contact the original author Jens Maurer
http://www.boost.org/users/people/jens_maurer.html
gives an email address.
Or does anyone else have a current address?
Paul
-----Original Message----- From: boost-bounces@lists.boost.org [mailto:boost-bounces@lists.boost.org] On Behalf Of Ross Levine Sent: 09 March 2009 05:46 To: boost@lists.boost.org Subject: Re: [boost] [random] Speed up binomial and poisson distributions
Hey Stephan,
I have a quick poisson generator based on Ahrens and Dieter's PD algorithm. I've been too busy to formally submit it. If you want to take a look it's here: http://dl.getdropbox.com/u/416909/poisson_distribution.hpp
If anyone knows how to submit this for review, please let me know.
Ross Levine
On Sat, Mar 7, 2009 at 5:04 AM, Paul A. Bristow <pbristow@hetp.u-net.com> wrote:
-----Original Message----- From: boost-bounces@lists.boost.org
[mailto:boost-bounces@lists.boost.org]
On
Behalf Of Stephan Sent: 06 March 2009 10:26 To: boost@lists.boost.org Subject: [boost] [random] Speed up binomial and poisson distributions
Hi,
I was wondering why the binomial distribution still isn't optimized for speed. If I look into the GSL as a comparison, they use much smarter and faster algorithms. As is, the bosst::random distributions are not usable for computational purposes to me, which is a pity since I very much like the approach using variate_generators and stuff...
I would be happy to contribute some faster code, but I don't know how exactly I should proceed for that... should I contact the maintainer directly?
Yes - I think you should - and report back here if you fail - someone else may be willing to take on the task
- or help *you* to take it on ;-)
Paul
--- Paul A. Bristow Prizet Farmhouse Kendal, UK LA8 8AB +44 1539 561830, mobile +44 7714330204 pbristow@hetp.u-net.com
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
AMDG Ross Levine wrote:
I sent Jens an email just now.
I have a style questions as well. First of all, this new poisson_distribution is policy-based (don't worry -- it's backwards-compatible with the old one; old code using the poisson_distribution will not need to be changed). Since poisson_distribution resides in the boost namespace, I didn't know if the policies should be placed there as well, or in the boost/random nested namespace. For now, they are in the boost/random/detail namespace and typedef'ed into the boost namespace.
It's dangerous to define public types in a detail namespace because of ADL.
They are named pd_fast, pd_small, and pd_legacy. However, it occurred to me that perhaps having policy classes in the boost namespace wasn't such a good idea, since these classes are not meant to be instatiated directly. At the very least, they could have private constructors and be friends with the poisson_distribution, but I still don't know if they ought to reside in the main boost namespace.
If anyone knows any rules for this, let me know. As it is there isn't much mention of this situation in the developer faq.
In addition, this header probably needs formal documentation. I am unaware of any stylistic guidelines I should follow.
They should go in namespace boost::random. In Christ, Steven Watanabe
Fixed. On Mon, Mar 9, 2009 at 9:54 PM, Steven Watanabe <watanabesj@gmail.com> wrote:
AMDG
Ross Levine wrote:
I sent Jens an email just now.
I have a style questions as well. First of all, this new poisson_distribution is policy-based (don't worry -- it's backwards-compatible with the old one; old code using the poisson_distribution will not need to be changed). Since poisson_distribution resides in the boost namespace, I didn't know if the policies should be placed there as well, or in the boost/random nested namespace. For now, they are in the boost/random/detail namespace and typedef'ed into the boost namespace.
It's dangerous to define public types in a detail namespace because of ADL.
They are named pd_fast, pd_small, and pd_legacy. However, it occurred to me that perhaps having policy classes in the boost namespace wasn't such a good idea, since these classes are not meant to be instatiated directly. At the very least, they could have private constructors and be friends with the poisson_distribution, but I still don't know if they ought to reside in the main boost namespace.
If anyone knows any rules for this, let me know. As it is there isn't much mention of this situation in the developer faq.
In addition, this header probably needs formal documentation. I am unaware of any stylistic guidelines I should follow.
They should go in namespace boost::random.
In Christ, Steven Watanabe
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
participants (5)
-
Neal Becker -
Paul A. Bristow -
Ross Levine -
Stephan -
Steven Watanabe