search

stan-dev / stan

Stan development repository. The master branch contains the current release. The develop branch contains the latest stable development. See the Developer Process Wiki for details.
https://mc-stan.org
BSD 3-Clause "New" or "Revised" License
2.56k stars 365 forks source link

Riemannian HMC #304

Closed betanalpha closed 8 years ago

betanalpha commented 10 years ago

There are three main updates necessary for an efficient implementation of Riemannian HMC (the algorithm is currently prototyped in feature/rhmc. @bob-carpenter and @syclik please added specific tasks under each update.

jgabry commented 8 years ago

This issue is very old and has had no updates. Is this still the issue for tracking RHMC or has that moved elsewhere? If the latter let's close this to reduce the number of open issues.

betanalpha commented 8 years ago

This is still active and waiting on some refactor decisions before being implemented in Stan.

On Mar 4, 2016, at 6:55 PM, Jonah Gabry notifications@github.com wrote:

This issue is very old and has had no updates. Is this still the issue for tracking RHMC or has that moved elsewhere? If the latter let's close this to reduce the number of open issues.

— Reply to this email directly or view it on GitHub.

bhargavvader commented 7 years ago

Is anyone working on RMHMC for Stan, at the moment? Is this project still being pursued?

syclik commented 7 years ago

The algorithm implementation is done! It has been for quite some time. Michael coded it up a long time ago.

Where we need help is extending automatic differentiation to 2nd and 3rd order for all the math functions. If you have calculus skills + some coding experience, we'd love to get some more help.

On Wed, Mar 8, 2017 at 8:47 AM, Bhargav Srinivasa notifications@github.com wrote:

Is anyone working on RMHMC for Stan, at the moment? Is this project still being pursued?

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/stan-dev/stan/issues/304#issuecomment-285044352, or mute the thread https://github.com/notifications/unsubscribe-auth/AAZ_F9fCgJKa2VE2QNmwEXgseFfb2ylnks5rjrFYgaJpZM4BHtKp .

bhargavvader commented 7 years ago

Is the completed code here - https://github.com/betanalpha/jamon ? I would love to help, but don't have the time right away - I was hoping to work on RMHMC for pymc3 as part of the GSoC 2017 program, but would be glad to help with the Stan implementation soon after!

betanalpha commented 7 years ago

No, literally coded up in Stan,

https://github.com/stan-dev/stan/blob/develop/src/stan/mcmc/hmc/hamiltonians/softabs_metric.hpp https://github.com/stan-dev/stan/blob/develop/src/stan/mcmc/hmc/hamiltonians/softabs_metric.hpp https://github.com/stan-dev/stan/blob/develop/src/stan/mcmc/hmc/hamiltonians/softabs_point.hpp https://github.com/stan-dev/stan/blob/develop/src/stan/mcmc/hmc/hamiltonians/softabs_point.hpp https://github.com/stan-dev/stan/blob/develop/src/stan/mcmc/hmc/integrators/impl_leapfrog.hpp

See also the associated tests. We just haven’t exposed it yet due to the potential fragility of the higher-order autodiff.

On Mar 8, 2017, at 11:07 AM, Bhargav Srinivasa notifications@github.com wrote:

Is the completed code here - https://github.com/betanalpha/jamon https://github.com/betanalpha/jamon ? I would love to help, but don't have the time right away - I was hoping to work on RMHMC for pymc3 as part of the GSoC 2017 program, but would be glad to help with the Stan implementation soon after!

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/stan-dev/stan/issues/304#issuecomment-285084158, or mute the thread https://github.com/notifications/unsubscribe-auth/ABdNlgv0MdQw4QcoaM_s0oqmVAStS4_uks5rjtI9gaJpZM4BHtKp.