I have no particular reason, I am using Metropolis in other parts of the code, where it is significantly faster than NUTS. Are you referring to NUTS as the “default sampler”? I have read some problems with the initialization - needed to use ADVI, which is allegedly at its best with large (and multidimensional) data (according to this post ).
dhajnes
5
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Bayesian regression using Gaussian mixture model | 0 | 558 | September 30, 2021 | |
| 2D Gaussian Mixture | 5 | 1265 | October 31, 2023 | |
| Problem with fitting Multivariate Mixture of Gaussians | 6 | 3588 | June 4, 2018 | |
| Mixture Gaussian mode dose not converges with high iterations | 0 | 297 | November 5, 2021 | |
| Updating priors for Gaussian Mixture Model | 5 | 1484 | May 26, 2022 |