Feature Request: Support Final Model Selection?

[billdenney]

One of the methods that I often use for generating models in R is to make a list of candidate models and then evaluate which is the best to select as the final model by one or more metrics. For that, I've written an AIC method that can compare lists of formula and put it into the PKNCA package. It's not a good fit for the underlying goal of the package, and I'd like to find a better home for the function and a few related functions (AIC.list, get.best.model, and get.first.model) as briefly described in billdenney/pknca#71.

Would you be interested in including functions (with tests) with a pull request that does the following?

• AIC.list: evaluate the AIC of a list of models.
  • AIC.logical: to handle NA in a list of models.
  • AIC.NULL: to handle NULL in a list of models.
• get.best.model: select the best model from a list of models by minimum AIC.
• get.first.model: select the first model that converged from a list of models.

[ewenharrison]

Potentially. Although I'm not enthusiastic about model selection on AIC alone.
Finalfit functions with metrics==TRUE includes an AIC for models. On RStudio, this can appear as a list of metrics in the environment tab. So just by scanning down the models already fitted, the best AIC can be seen directly. Do you think the function adds significantly to that?
Many thanks for the interest. It is greatly appreciated.

[billdenney]

Sorry for my long silence on this issue. As I thought about it more, I agree that it doesn't feel like a great fit for finalfit.

Thanks again for your work on the package!