I am using pyro.optim.ClippedAdam and have tried clip_norm = 0.00001, 1.0, 10 and a bunch of values in between, but the gradients are always regardless and don’t appear to change when I change the clip_norm.
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 so I tried this BTYD model below and I got 879 divergences! So I tried to do use the reparam config with LocScaleReparam (0) but that would not work on the Gamma and Inverse Gamma functions. from numpyro.handlers imp…
Q4: running on a single GPU (K80) or even 4GPUs does not improve the running speed. Do you catch a reason for that? Thanks for your advises. PS: I have create an issue on Pyro github but I am not sure that it is the correct way. Hope that you can close the issue in case it was not the right way to ask the question. fehiepsi ...
Hello, first off, amazing job on Pyro! Major kudos 🙂 How do I sample from the posterior predictive for an SVI-trained model efficiently? At the moment, I sample a guide trace for each desired posterior predictive sample, replay the model with the guide trace, and sample once from it, like this: ppc = [] dummy_obs = torch.zeros((1,self.D)) for sample in range(n_samples): guide_trace = pyro ...
Hi there, I want to implement a WAIC (Watanabe–Akaike information criterion) to help me on model evaluation & seletion. For this purpose, I need to calculate a log pointwise predictive density (lppd) as follows: This involves a density for each posterior sample and each observation. And I noticed that there is a runtime utility log_density() in NumPyro. My question is, can I use it, first ...
"The value argument must be within the support ... - Pyro Discussion Forum