Initializing large correlation matrices is numerically tricky, and Stan’s default initialization of these matrices is numerically fragile. A method exists to initialize them more robustly, and is commonly called the onion method. Here’s a good place to begin reading: Using the onion method in occupancy model hierarchical model - #2 by spinkney
jsocolar
2
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Unable to Initialize Large Correlation Matrix In PyStan | 1 | 834 | March 1, 2022 | |
| Initialization between (-2, 2) failed after 100 attempts | 4 | 488 | May 22, 2023 | |
| Failure to start because of initial values | 16 | 3624 | July 31, 2017 | |
| Lkj_corr_lpdf: Correlation matrix is not positive definite | 9 | 1284 | July 12, 2023 | |
| Underestimating correlation coefficients with LKJ prior | 24 | 7190 | September 15, 2020 |