jnothman
Help user interpret param columns of cv_results_
Because the user no longer specifies the grid by full names, results are then presented with param names that are unfamiliar and which will vary with
The user needs help dereferencing the parameters. What is a good API for this?
global_to_local(gscv, double_underscored_name) -> (local_est, local_param)
local_to_global(gscv, local_est) -> global_param
what if set_grid() accepted an optional name parameter, which would be used when describing the cv_results_ params? If the name parameter isn't supplied, the code could default to exactly how it works today (less work in describing the parameter, more difficulty in interpreting the results).
I don't mind the idea of having some kind of name_ parameter to set_grid, but we would have to have a custom GridSearchCV.fit implementation in order to do the rename in cv_results_, so there may be substantial maintenance overhead and additional complexity for the user who now has to appreciate that sklearn.model_selection.GridSearchCV and searchgrid.GridSearchCV are distinct classes. But it might be justifiable
Ah, I see. Having a custom GridSearchCV would be a good deal more overhead. I'm not very familiar with sklearn's GridSearchCV details under the hood, so I was hoping that this wouldn't need to reimplement anything. Let me familiarize myself more with that part of the sklearn codebase a little more.
It wouldn't need to copy the GridSearchCV code, just use inheritance, so it's not a huge deal, it just might confuse users... Maybe worthwhile.