'i-esim information criterion' in Model Averaging

Questions
1

Would anyone advise what ‘i-esim’ information criterion means?

As I read PyMC3 document on model averaging, I encountered this unfamiliar term. (https://docs.pymc.io/notebooks/model_averaging.html#Pseudo-Bayesian-model-averaging)
The sentence is under the formula of the link above.

Where dICidICi is the difference between the i-esim information criterion value and the lowest one. Remember that the lowest the value of the IC, the better. We can use any information criterion we want to compute a set of weights, but, of course, we cannot mix them.

Thank you,
Shawn

1 Like
2

I think you have found a typo, it should just be i-th superscript th

dIC shows the difference between that row’s IC and the best IC out of all the rows in the compare table.

1 Like
3

Hi Shawn,
Feel free to report it as a GH issue, or even better as a PR :slight_smile:
This NB should also probably be updated – plots and diagnotics are now delegated to ArviZ, and the default scale is now log instead of deviance (which means a higher IC is better).

1 Like
4

Thanks for the warm replies for the first question from PyMC newbie!

1 Like