Models: maximum likelihood vs Bayesian
Closed this issue · 5 comments
We need some additional text on the difference between maximum likelihood models and Bayesian models; e.g., how to get error bars from the latter.
Where would you include that? Notebook 03 - MSM estimation and validation?
The conceptual difference should definitely go into 03, but I would like to show how it is used to get error bars in 04 and 06.
I think it makes sense to mention both in 03 and highlight how you get errors on time-scales and ck-tests. In 04 we can show sample_mean, sample_conf and sample_std for example in the context of the MFPTs to show how the Bayesian model allows us quantify errors in these quantities -- I use these functions in 06, but I think it would make sense to also mention them in an earlier notebook. Thoughts?
Good point to focus on CK+ITS error bars in NB04. We only need to adapt the text then.
Yes, I'll try to explain it at examples that already use Bayesian MSMs and show how to explicitly get the estimates outside of timescales_msm(dtrajs, erros='Bayes') using the functions Simon mentioned. I'd rather avoid explaining Bayesian HMMs though.