tabensemb.trainer.Trainer.get_approx_cv_leaderboard#
method
- Trainer.get_approx_cv_leaderboard(leaderboard: DataFrame, save: bool = True) Tuple[DataFrame, DataFrame][source]#
Calculate approximated averages and standard errors based on
cross_validation()results in the folderself.project_root/cv.- Parameters:
- leaderboard
A reference leaderboard to be filled by avg and std, and to sort the returned DataFrame.
- save
Save returned results locally with names “leaderboard_approx_mean.csv” and “leaderboard_approx_std.csv”
- Returns:
- pd.DataFrame
Averages in the same format as the input
leaderboard. There is an additional column “Rank”.- pd.DataFrame
Standard errors in the same format as the input
leaderboard. There is an additional column “Rank”.
Notes
The returned results are approximations of the precise leaderboard from
get_leaderboard. Some metrics like RMSE may be different because data-point-wise and cross-validation-wise averaging are different.