Reason codes: the variables that took the most points from each row
Source:R/apply.R
scr_reasons.RdFor each row of newdata, the k variables whose contribution in points
fell furthest below the reference. The reference is the mean points of
the variable on the training population ("mean", the Regulation B safe
harbor referenced to the average) or the maximum points of the variable
("max"). Only applies to the additive scorecard; a tree challenger has
no reason codes.
Usage
scr_reasons(x, newdata, k = 4L, reference = c("mean", "max"))Arguments
- x
An object from
scr_scorecard().- newdata
New table.
- k
Number of reasons per row.
- reference
"mean"(default) or"max".
Value
A data.table with reason_1 ... reason_k (variable names)
and shortfall_1 ... shortfall_k (points below the reference).
Details
Under higher_is_riskier the shortfall is measured the other way round:
the reasons are the variables that added the most points.
See also
Other production:
predict.scr_align(),
scr_apply(),
scr_export(),
scr_monitor(),
scr_monitoring_plan(),
scr_sql()
Examples
cfg <- scr_config(verbose = FALSE, nthread = 1, use_ranger = FALSE,
xgb_rounds = 60, n_boot = 20)
res <- scr_select(scr_demo, "default", config = cfg, drop = "id",
date_col = "ref_date")
sc <- scr_scorecard(res)
scr_reasons(sc, head(scr_demo, 5), k = 3)
#> reason_1 shortfall_1 reason_2 shortfall_2 reason_3 shortfall_3
#> <char> <num> <char> <num> <char> <num>
#> 1: vl_score_01 25.197857 vl_score_02 17.828571 vl_score_05 9.054286
#> 2: vl_score_04 9.052857 vl_late 5.716071 ds_channel 5.387857
#> 3: vl_late 13.716071 vl_score_07 7.605357 ds_channel 5.387857
#> 4: ds_region 13.807143 ds_band 11.932500 vl_score_02 11.828571
#> 5: vl_score_02 17.828571 vl_score_05 9.054286 vl_score_01 5.197857