I am learning Bayesian Model Updating and using DRAM sampling method to solve posterior PDF, I have trouble in DRAM method. The issue is : The matlab program I used is open source you uploaded, then I changed it according to my case(total 6 parameters to be updated),I have verified my program for 2 parameters, it works well. However, when the number of parameters are 6, samples are not taken to convergence. The prior I used is gaussian distribution with infinit variance (so the prior can be seen as noninformative prior), I also set parameters interval, initial value, qcov, drscale and other DRAM setting, but results are still awful.Does anybody do DRAM sampling method before? Thanks!
Stop using noninformative priors, no wonder nothing converges.
You should pecify your model in Stan, and I believe there is a package for matlab. Then you can gather samples and there will be multiple convergence diagnostics you can monitor afterwards.
- Badges
- Science topic
- Similar topics
- Computer Science
- Program