2022年2月21日 · Hi all, I have a model which I successfully trained with SVI and the results are good. I want to add the subsampling option to the numpyro.plate. The relevant part of the model is given bellow. However, I have a problem with the shape of the obs variable when I draw the prior/posterior distribution from the model. What I get for the prior_predictions['obs'] has the …

2020年12月3日 · I find Forecasting III: hierarchical models — Pyro Tutorials 1.8.4 documentation do not cover a problem : nan in real data . For example : in 2010, there was only 48 station in 2011, 3 station was closed and 5 new opened, 50 stations . in 2012, 2 station was closed in 2011 reponed … For another example : My data is salecount of various products in many stores . I …

2022年11月24日 · I’m seeking advice on improving runtime performance of the below numpyro model. Model Description: I have a dataset of L objects. For each object, I sample a discrete variable c and eight continuous variables – s , h and six parameters theta_i which determine an analytical function (defined by the method dst). This function is fit to observed data points, one …

2023年8月1日 · I’ve read through the Pyro docs on tensor shapes, but I’m still learning the ropes of Pyro, so I might be missing something simple… When I use the print_shapes argument for the HMM script without vectorized particles, it looks like the batch dims are: 16, 16, 8, 51 , which I believe correspond to: <hidden dim at time t>, <hidden dim at t ...

2020年7月21日 · Following up on @fritzo’s responses here and here on how to handle missing (MCAR) data. Sequential observations would be straight-froward but would hamper runtime too much. poutine.mask is great and I’ve been able to get it to work filtering out missing values from leaf nodes, as in the linked examples. What I’m currently stuck on, however, is how to deal …

2022年1月16日 · Here’s some code I’ve written, but I’m struggling to make sense of how to fix it. There’s some observed data from two countries, from two cities each. So I’d like to make a hierarchical model with two levels: each city can have its own mean and std each city within the same country shares hyperpriors each country shares global hyperpriors Here’s a complete …

2022年2月7日 · Hello, I’m trying to setup a demo notebook on Google Collab (K80) but I face the problem of slow SVI. I try a AutoMultivariateNormal with Adam optimizer and loss=Trace_ELBO(num_particles=1) with 5_000 steps. It seems that it will take 6h or so. Is there some way to speed up the optimization? Thanks

2024年11月1日 · Hi, I’m working on a model where the likelihood follows a matrix normal distribution, X ~ MN_{n,p} (M, U, V). I’m using conjugate priors: M ~ MN U ~ Inverse Wishart V ~ Inverse Wishart As a result, I believe the posterior distribution should also follow a matrix normal distribution. Is there a way to implement the matrix normal distribution in Pyro? If I replace the …

2024年3月29日 · Hi, from previous posts, setting some of the priors to tensors on cuda should allow me to run pyro on gpu. but doesn’t seem to work for me. could you pl shelp? import pyro from pyro.nn import PyroSample import torch from torch import nn from pyro.nn import PyroModule from pyro.infer.autoguide import AutoDiagonalNormal from pyro.infer import SVI, …

2023年12月6日 · Hello, I am new to pyro and BNN and I am still trying to properly understand the duality prior, posterior. So far, I understood that the prior is supposed to represent the knowledge I have, my intuitions BEFORE learning, so they are often chosen as isotropic Gaussians. Yet as the prior has a role in the prediction function, a bad prior could result in a bad prediction, thus you …