Re: Implementations of Markov chain Monte Carlo methods?

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
6 messages Options
Reply | Threaded
Open this post in threaded view
|

Re: Implementations of Markov chain Monte Carlo methods?

Ross Boylan


On Wednesday, April 10, 2013 8:44:18 PM UTC-7, Viral B. Shah wrote:
I just stumbled upon some analysis of Stan that Stefan had done a while back. Given that it is C++ templated code, it is not convenient to call from julia. However, Stefan's analysis may help convince someone to either translate or re-implement these methods in Julia:

For the record (I just found this thread after noticing that MCMC was a focus area for Julia stats), stan is callable from R and python, and so it would surely be possible to do so from Julia.  stan has been somewhat refactored over the last year or so, which should make it easier to slice and dice.  Note that "easier" != "easy."

I've also noticed the stan developers (who are statisticians) keep adding new algorithms and tweaking old ones, which seems to me an argument to use their work rather than reimplement it.

I've used stan extensively from R and looked into packaging it for Debian, and so have some sense of its internal structure. Stan is pretty heavy weight; it uses a bunch of other libraries and even stan users need to have a compiler, since the typical work flow is that stan translates a user model to C++ code, which is then compiled (I believe, though am not sure, the the "symbolic" differentiation happens at compile time) and the resulting executable does the actual model estimation.

Ross Boylan

--
You received this message because you are subscribed to the Google Groups "julia-stats" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.
Reply | Threaded
Open this post in threaded view
|

Re: Implementations of Markov chain Monte Carlo methods?

John Myles White
Hi Ross,

Please look into Stan.jl: https://github.com/goedman/Stan.jl

This thread is very old and doesn't reflect any of the work that's happened since April 2013.

 -- John

On Aug 20, 2014, at 3:45 PM, [hidden email] wrote:



On Wednesday, April 10, 2013 8:44:18 PM UTC-7, Viral B. Shah wrote:
I just stumbled upon some analysis of Stan that Stefan had done a while back. Given that it is C++ templated code, it is not convenient to call from julia. However, Stefan's analysis may help convince someone to either translate or re-implement these methods in Julia:

For the record (I just found this thread after noticing that MCMC was a focus area for Julia stats), stan is callable from R and python, and so it would surely be possible to do so from Julia.  stan has been somewhat refactored over the last year or so, which should make it easier to slice and dice.  Note that "easier" != "easy."

I've also noticed the stan developers (who are statisticians) keep adding new algorithms and tweaking old ones, which seems to me an argument to use their work rather than reimplement it.

I've used stan extensively from R and looked into packaging it for Debian, and so have some sense of its internal structure. Stan is pretty heavy weight; it uses a bunch of other libraries and even stan users need to have a compiler, since the typical work flow is that stan translates a user model to C++ code, which is then compiled (I believe, though am not sure, the the "symbolic" differentiation happens at compile time) and the resulting executable does the actual model estimation.

Ross Boylan

--
You received this message because you are subscribed to the Google Groups "julia-stats" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "julia-stats" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.
Reply | Threaded
Open this post in threaded view
|

Re: Implementations of Markov chain Monte Carlo methods?

Ross Boylan
On Wed, Aug 20, 2014 at 03:59:50PM -0700, John Myles White wrote:
> Hi Ross,
>
> Please look into Stan.jl: https://github.com/goedman/Stan.jl
>
> This thread is very old and doesn't reflect any of the work that's happened since April 2013.
>
>  -- John
Thanks; that looks like exactly the approach I was thinking of.

I have a meta question as someone new to julia and new to using github
as more than a code repository: why doesn't there seem to be any
mention of that recent work on this list?  Searching for stan on this
list turned up the thread I mentioned and another, newer one from
12/2013 that mentioned stan in passing, in the context of
reimplementing its algorithms in julia.

Ross

--
You received this message because you are subscribed to the Google Groups "julia-stats" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.
Reply | Threaded
Open this post in threaded view
|

Re: [julia-dev] Implementations of Markov chain Monte Carlo methods?

Kevin Squire
I can't explain why it didn't turn up while searching, but here are a couple of relevant links:


Cheers,
   Kevin

On Thursday, August 21, 2014, Ross Boylan <[hidden email]> wrote:
On Wed, Aug 20, 2014 at 03:59:50PM -0700, John Myles White wrote:
> Hi Ross,
>
> Please look into Stan.jl: https://github.com/goedman/Stan.jl
>
> This thread is very old and doesn't reflect any of the work that's happened since April 2013.
>
>  -- John
Thanks; that looks like exactly the approach I was thinking of.

I have a meta question as someone new to julia and new to using github
as more than a code repository: why doesn't there seem to be any
mention of that recent work on this list?  Searching for stan on this
list turned up the thread I mentioned and another, newer one from
12/2013 that mentioned stan in passing, in the context of
reimplementing its algorithms in julia.

Ross

--
You received this message because you are subscribed to the Google Groups "julia-stats" group.
To unsubscribe from this group and stop receiving emails from it, send an email to <a href="javascript:;" onclick="_e(event, &#39;cvml&#39;, &#39;julia-stats+unsubscribe@googlegroups.com&#39;)">julia-stats+unsubscribe@....
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "julia-stats" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.
Reply | Threaded
Open this post in threaded view
|

issues and discussion lists (was Re: Implementations of Markov chain Monte Carlo methods?)

Ross Boylan
On Thu, Aug 21, 2014 at 12:39:07AM -0700, Kevin Squire wrote:
> I can't explain why it didn't turn up while searching, but here are a
> couple of relevant links:
>
> https://groups.google.com/forum/m/#!topic/stan-users/p67ythJ1J0M
That didn't turn up since I searched julia-stats only.

> https://github.com/JuliaStats/MCMC.jl/issues/45
Ditto.

How does one become aware of an issue?  In this particular case the
thread in stan-users had a reference to the issue, but is there, for
example, a way to get notices of new issues (or ones with certain
keywords)?

When is it better to raise something as an issue vs a message to a
mailing list/google group?

Ross

>
> Cheers,
>    Kevin
>
> On Thursday, August 21, 2014, Ross Boylan <[hidden email]> wrote:
>
> > On Wed, Aug 20, 2014 at 03:59:50PM -0700, John Myles White wrote:
> > > Hi Ross,
> > >
> > > Please look into Stan.jl: https://github.com/goedman/Stan.jl
> > >
> > > This thread is very old and doesn't reflect any of the work that's
> > happened since April 2013.
> > >
> > >  -- John
> > Thanks; that looks like exactly the approach I was thinking of.
> >
> > I have a meta question as someone new to julia and new to using github
> > as more than a code repository: why doesn't there seem to be any
> > mention of that recent work on this list?  Searching for stan on this
> > list turned up the thread I mentioned and another, newer one from
> > 12/2013 that mentioned stan in passing, in the context of
> > reimplementing its algorithms in julia.
> >
> > Ross

--
You received this message because you are subscribed to the Google Groups "julia-stats" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.
Reply | Threaded
Open this post in threaded view
|

Re: issues and discussion lists (was Re: Implementations of Markov chain Monte Carlo methods?)

Kevin Squire
On Thu, Aug 21, 2014 at 1:00 AM, Ross Boylan <[hidden email]> wrote:
On Thu, Aug 21, 2014 at 12:39:07AM -0700, Kevin Squire wrote:
> I can't explain why it didn't turn up while searching, but here are a
> couple of relevant links:
>
> https://groups.google.com/forum/m/#!topic/stan-users/p67ythJ1J0M
That didn't turn up since I searched julia-stats only.

> https://github.com/JuliaStats/MCMC.jl/issues/45
Ditto.

How does one become aware of an issue?  In this particular case the
thread in stan-users had a reference to the issue, but is there, for
example, a way to get notices of new issues (or ones with certain
keywords)?

I'm aware of it because I follow the different mailing lists pretty closely (and have for a while), but I found those just though a normal Google search.

If you have a github account, you can follow a repo by going to it and clicking the "Watch" button, and you can follow individual issues in a similar manner.  If you're the owner of (or belong to the group that is owner of) a repository, you're subscribed automatically.

As an aside, since you're in biostats, you might want to check out the biojulia-dev group and the BioJulia group on github.

When is it better to raise something as an issue vs a message to a
mailing list/google group?

Issues are generally for bugs, or sometimes feature requests.  It's often good to query on the mailing list first--julia-users is for basic questions, julia-stats for stats-related questions, biojulia-dev for bio related, etc.

Cheers,
   Kevin



> On Thursday, August 21, 2014, Ross Boylan <[hidden email]> wrote:
>
> > On Wed, Aug 20, 2014 at 03:59:50PM -0700, John Myles White wrote:
> > > Hi Ross,
> > >
> > > Please look into Stan.jl: https://github.com/goedman/Stan.jl
> > >
> > > This thread is very old and doesn't reflect any of the work that's
> > happened since April 2013.
> > >
> > >  -- John
> > Thanks; that looks like exactly the approach I was thinking of.
> >
> > I have a meta question as someone new to julia and new to using github
> > as more than a code repository: why doesn't there seem to be any
> > mention of that recent work on this list?  Searching for stan on this
> > list turned up the thread I mentioned and another, newer one from
> > 12/2013 that mentioned stan in passing, in the context of
> > reimplementing its algorithms in julia.
> >
> > Ross

--
You received this message because you are subscribed to the Google Groups "julia-stats" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "julia-stats" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [hidden email].
For more options, visit https://groups.google.com/d/optout.