search

biometrician / abe

An R package for Augmented Backward Elimination
GNU General Public License v3.0
3 stars 0 forks source link

#17: summary of model selection frequencies #17

Closed biometrician closed 1 year ago

biometrician commented 1 year ago

Currently summary(abe_object) reports under $model.rel.frequencies a list of all selected models with their frequencies. Is it possible to change the list so that it looks similar to Table 6 in the Heinze et al 2018 paper.

Additionally, maybe put a star to the selected model in the list.

biometrician commented 1 year ago

Is it okay, if I ask Gregor to do this?

rokblagus commented 1 year ago

We should wait until we decide what to do with the arguments of abe.resampling, this would be best discussed in person.

rokblagus commented 1 year ago

Note that print.abe, pie.abe and plot.abe depend on the summary.abe function, so this functions should not be changed much!

biometrician commented 1 year ago

Since the table with the frequencies of different selected models should be reported, we could add this to the print.abe function. As with plot.abe where different plots can be requested. We could say print.abe( ..., type = "models") gives a nice table similar to the one in Heinze et al 2018 Table 6. print.abe( ..., type = "coefficients") would be the default, which is the current table of estimates. What do you say?

gregorsteiner commented 1 year ago

I just added this. The print function now has a "models" type which produces a table similar to the one in Heinze et al. (2018). let me know if you have any suggestions for improvement.

biometrician commented 1 year ago

Looks nice! Thx. Could you include a parameter so that the analyst can modify the number of reported models?

Either we have 2 parameters, e.g., model.cum.freq and models.n were the user can override the default and either select the cumulative frequency or the number of models. Alternatively, we could have just one parameter and if numbers are <= 1 then it is assumed cumulative frequencies are meant and if numbers are >1 then the number of models is requested. However, at the moment I have no idea how to name this parameter nicely.

(If nothing is added, then the default cum. freq of 0.8 or the top 20 models whichever list is shorter.)

gregorsteiner commented 1 year ago

I implemented the one parameter version. I think I prefer having just one extra argument. There is now a parameter models.n, which can either be the absolute number of models to return, or the cumulative frequency to attain. I added some documentation explaining this, since this may not be super intuitive.

biometrician commented 1 year ago

This is really nice! I also prefer the version with just one additional parameter. Thanks a lot!