I am confused about how arviz.compare() reports the 'rank'. In the documentation it says:
rank: The rank-order of the models. 0 is the best.
Is this best decided taking into account both predictive accuracy as well as simplicity OR is it only based on the predictive accuracy of the models?
And the same question applies to the weight as well. Is the weight given to the models is based on both predictive accuracy as well as simplicity OR is it only based on the predictive accuracy of the models?
Here is the link to the documentation I am referring to.