Hi @matsuo_basho
I’m guess you mean ‘sample’ rather than build? A few quick observations:
- you are importing PyMC 3, which is quite old now. The latest is version 5. So you could try updating your version.
- I would recommend starting with the default
target_acceptparameter. Ie, try just removing this argument as it’s currently set very high. - Exponential priors on standard deviations might not be suitable. If you end up with zero standard deviations then this won’t work very well. So you could try
HalfNormalorGammadistributions.