Bayesian hmm

Hi guys,
I have an hmm model that works perfect with auto delta guide.
I try to use auto diagonal normal to the same model and i can see that prediction are not good
when i try to plot my output point I can see more clusters that the number I was expected.
my guide call looks like :
guide = AutoDiagonalNormal(poutine.block(model,expose=[“probs_x”,“probs_y”,“probs_z”,“probs_kappa1”,“probs_kappa2”,“probs_kappa3”,“probs_y”,“probs_a”,] ), init_loc_fn = init_to_sample)

To see if my model works I try to remove some of latent variable sand I get the result I expect for the given data.
guide = AutoDiagonalNormal(poutine.block(model,expose=[“probs_x”,“probs_y”,“probs_z”,“probs_kappa1”,“probs_kappa2”,“probs_a”,] ), init_loc_fn = init_to_sample).

It seems that when I add more variables corresponding Bayesian model do not perform I expect.
Do you have any idea about that ?

Thank you in advance for your help.

Please share model code with us if you’d like us to help. I can’t figure out what the problem is if I don’t see what your model is, particularly if the issues are statistical (e.g., your model executes fine but has poor inference behavior).

Also, please write code with backticks (`) , like

guide = ...