search

stefanradev93 / BayesFlow

A Python library for amortized Bayesian workflows using generative neural networks.
https://bayesflow.org/
MIT License
297 stars 45 forks source link

how to determine the number of layers #107

Closed Shuwan-Wang closed 10 months ago

Shuwan-Wang commented 10 months ago

Hi Stefan et al.,

I am curious that how did you determine the number of layers for the different models?

In the simulation studies, you mentioned like 6-10 ACBs should be enough for moderate model setup. So instead of trying different #of ACBs and keep stacking up, do you have any suggestions/better way?

Thanks! Best, Shuwan

stefanradev93 commented 10 months ago

Hey Shuwan,

There is no established systematic way. Bayesian optimization would be one, but it may be an overkill to just testing out a few variants. Stefan