One row per scr_config() key, with the stage it acts on, the default
value and what it controls.
See also
Other configuration:
scr_config(),
scr_presets(),
scr_verbose()
Examples
head(scr_config_keys(), 8)
#> key stage default
#> 1 preset 0 moderate
#> 2 objective 0 risk
#> 3 seed 0 2203
#> 4 nthread 0 2
#> 5 verbose 0 TRUE
#> 6 oot_date_col 0 NULL
#> 7 holdout_ratio 0 0.30
#> 8 special_values 1 -999
#> description
#> 1 Tightness of the selection funnel
#> 2 Convention: target=1 is an undesirable (risk) or desirable (propensity) event
#> 3 Seed of everything random
#> 4 Parallel workers (binning by column, bootstrap, CSI)
#> 5 Progress messages
#> 6 Date column of the out-of-time cut
#> 7 Target hold-out fraction
#> 8 Business sentinels
scr_config_keys(stage = 5)
#> key stage default
#> 1 base_score 5 600
#> 2 base_odds 5 50
#> 3 pdo 5 20
#> 4 direction 5 NULL
#> 5 align_method 5 regression
#> 6 align_bands 5 10
#> 7 points_style 5 base_plus_deviation
#> 8 points_round 5 TRUE
#> 9 challenger 5 NULL
#> 10 max_abs_coef 5 15
#> 11 n_boot 5 200
#> 12 ci_level 5 0.95
#> 13 score_groups 5 10
#> description
#> 1 Reference score of the scale
#> 2 Odds at the reference score (in the orientation of direction)
#> 3 Points to double the odds
#> 4 Scale direction; NULL derives it from objective
#> 5 regression (bands + ln(odds) ~ raw) or direct
#> 6 Bands of the alignment regression
#> 7 Points: base + deviation, or distributed intercept
#> 8 Round points per bin
#> 9 Tree challenger (xgboost/lightgbm), without points
#> 10 Maximum absolute glm coefficient tolerated
#> 11 Bootstrap CI resamples (always computed)
#> 12 CI level
#> 13 Bands of the score gains