truncated DP mixture model example

[dustinvtran]

(Elham Azizi asks about DP mixture models for use in some of her work with Dana Pe’er in computational biology.)

part of my reply below:

DPs in their full infiniteness aren’t possible in Edward for similar reasons as Stan. It requires lazy evaluation to bake in a generative process with an infinite number of latent variables but whose output only uses a finite number of latent variables. We’re still working on that regard.

Truncated DP mixture models are definitely possible, e.g., setting the maximum number of stick breaks a priori. And it seems like that’s what your work uses. I imagine Edward is the fastest for inferring it in practice because all other systems out there use MCMC or SMC. And Edward will handle data subsampling and things like GPU support or distributed training if you want to take advantage of them.

Would be great if someone gets a truncated DP mixture model up and running!

[AustinRochford]

I have a PyMC3 truncated DP mixture model example currently via MCMC here. It may be possible to adapt this into an Edward example by swapping out some of the Theano-specific bits, if necessary. I'd be happy to give this a shot once I get back from an internet-less vacation next week.

[dustinvtran]

@AustinRochford: yeah that would be fantastic if you could.

[elhamazizi]

That would be really helpful, thanks! @AustinRochford

[AlexPiche]

@AustinRochford, I was wondering if you were working on the issue? If yes, I would be happy to help you out. I am familiar with DP, nested DP and hierarchical DP mixtures.

[AustinRochford]

@AlexPiche I had done some work on it, then my day job got in the way. I am hoping to pick it up again in the next two or three weeks

[jengelman]

@AustinRochford did you ever end up picking this up again? I'm doing some work on nonparametrics and would be happy to contribute.

[AustinRochford]

I did not get a chance to get very far on this, so feel free to pick it up if you're so inclined!

[elhamazizi]

Hi Dustin, Hope all is well! I was wondering if you know if anyone ever implemented the DP mixture model in Edward. We have another (comp bio) model that we'd like to implement in Edward if possible. Thanks! Best, Elham

…

On Thu, Apr 6, 2017 at 12:47 PM Joshua Engelman ***@***.***> wrote: @AustinRochford <https://github.com/AustinRochford> did you ever end up picking this up again? I'm doing some work on nonparametrics and would be happy to contribute. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#183 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHrctsuQhJbu6TDP1uyoOBUpZ2iwFuBMks5rtRczgaJpZM4JSJJt> .

[dustinvtran]

hi @elhamazizi. The answer is an unfortunate no. I believe there's a truncated DP model if you search the Forum. But we never got efficient inference to work on it(?).

[elhamazizi]

Hi Dustin, Thanks, the truncated DP seems to work for now. Another question we have is whether you're planning to incorporate the Inverse-Wishart distribution into Edward? Best, Elham

…

On Tue, Feb 13, 2018 at 12:15 PM Dustin Tran ***@***.***> wrote: hi @elhamazizi <https://github.com/elhamazizi>. The answer is an unfortunate no. I believe there's a truncated DP model if you search the Forum. But we never got efficient work on it(?). — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#183 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHrcth00OtomewELpmtX_D13hBtHN8Gtks5tUcNHgaJpZM4JSJJt> .

[dustinvtran]

You can try submitting a feature request for that in TF's Github issues.

…

On Feb 16, 2018, at 10:17 AM, elhamazizi ***@***.***> wrote: Hi Dustin, Thanks, the truncated DP seems to work for now. Another question we have is whether you're planning to incorporate the Inverse-Wishart distribution into Edward? Best, Elham On Tue, Feb 13, 2018 at 12:15 PM Dustin Tran ***@***.***> wrote: > hi @elhamazizi <https://github.com/elhamazizi>. The answer is an > unfortunate no. I believe there's a truncated DP model if you search the > Forum. But we never got efficient work on it(?). > > — > You are receiving this because you were mentioned. > > > Reply to this email directly, view it on GitHub > <#183 (comment)>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/AHrcth00OtomewELpmtX_D13hBtHN8Gtks5tUcNHgaJpZM4JSJJt> > . > — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#183 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACc2i2PX7fjvll4DdYIQJqIrZurT0-S0ks5tVcYcgaJpZM4JSJJt>.