Hello again, Related post: Batch processing Pyro models so cc: @fonnesbeck as I think he’ll be interested in batch processing Bayesian models anyway. I want to run lots of numpyro models in parallel. I created a new post because: this post uses numpyro instead of pyro I’m doing sampling instead of SVI I’m using Ray instead of Dask that post was 2021 I’m running a simple Neal’s funnel ...
Hi everyone, I am very new to Numpyro and hierarchical modeling. Recently I have been trying to build a hierarchical model where I have a hyper-prior (theta_group) which should be centered around 1 and strictly positive. There is another prior (theta_part) which should be centered around theta_group. I am trying to use LogNormal as priors for both: theta_group = numpyro.sample("theta_group ...
Do you mean using Predictive function as discussed in this pyro-forum-question. I just used the Predictive function without providing posterior distribution and I am able to generate samples for “obs” site .
Model and guide shapes disagree at site ‘z_2’: torch.Size ( [2, 2]) vs torch.Size ( [2]) Anyone has the clue, why the shapes disagree at some point? Here is the z_t sample site in the model: z_loc here is a torch tensor wi…
i see. this would appear to be a bug/unsupported feature. if you like, you can make a feature request on github (please include a code snippet and stack trace). however, in the short term your best bet would be to try to do what you want in pyro, which should support this.
I am running NUTS/MCMC (on multiple CPU cores) for a quite large dataset (400k samples) for 4 chains x 2000 steps. mcmc.run actually ran until the end, but then died with an out-of-memory exception; I assume upon trying to gather all results. (There might be some unnecessary memory duplication going on in this step?) Are there any “quick fixes” to reduce the memory footprint of MCMC? For ...
Hi all, I’ve read a few posts on the forum about how to use GPU for MCMC: Transfer SVI, NUTS and MCMC to GPU (Cuda), How to move MCMC run on GPU to CPU and Training on single GPU, but there are a few questions I still have on how to get the most out of numpyro. There is also this blog post comparing MCMC sampling methods on GPU, and although the model is built in pymc, it uses numpyro ...
pyro provides access to the pytorch schedulers, and the pyro ClippedAdam also has a specific learning rate decay parameter. I can not find anything of the sort in numpyro, however, or any example that does this?
