Funnel of Hell - STAN models

[drbenvincent]

Current status, before reparamaterization:

Non-hierarchical models seem to have no chain pathologies, so the problem lies with the hierarchical aspect of the model.

• hyperbolic + magnitude effect models Separate: ok. Mixed & Hierarchical: Rhat ok, but chain pathologies
• hyperbolic Separate: ok. Mixed: problematic Rhat. Hierarchical: seems ok.
• exponential Separate: ok. Mixed & Hierarchical: chain pathologies.
• not appropriate for non-parametric it seems
• exp-power model is different. Very very bad. Doesn't seem to converge for separate model, and stan exits for the mixed model. (Full hierarchical model not implemented yet)

Implement updates for the STAN models:

• [ ] hyperbolic + magnitude effect models
• [ ] hyperbolic
• [ ] exponential
• not appropriate for non-parametric it seems
• [ ] exp-power model

[drbenvincent]

Abandoning use of STAN in this toolbox. See darc-analysis-python for an implementation with more modern samplers.