scrimp_vars: Score imputations for specific variables
In bcjaeger/ipa: Imputation for Predictive Analytics

Description Usage Arguments Value Note Examples

If you want to evaluate how accurately an imputation procedure fills in missing values, scrimp_vars can help. Generally, scrimp_vars only applies to artificial situations where you ampute your data (i.e., make missing values), then impute it. For a more general imputation validation procedure, see scrimp_mdl.

scrimp_vars(
  data_imputed,
  data_missing,
  data_complete,
  miss_indx = NULL,
  fun_ctns_error = yardstick::rsq_trad_vec,
  fun_intg_error = yardstick::rsq_trad_vec,
  fun_bnry_error = yardstick::kap_vec,
  fun_catg_error = yardstick::kap_vec
)

`data_imputed`	an imputed data frame.
`data_missing`	the unimputed data frame.
`data_complete`	a data frame containing the 'true' values that were 'missing'.
`miss_indx`	an object returned from the mindx function applied to `data_missing`.
`fun_ctns_error`	a function that will evaluate errors for continuous variables. Continuous variables have type `double`. Default is to use R-squared (see `yardstick::rsq()`).
`fun_intg_error`	a function that will evaluate errors for integer valued variables. Default is to use R-squared (see `yardstick::rsq()`).
`fun_bnry_error`	a function that will evaluate errors for binary variables (i.e., factors with 2 levels). Default is to use kappa agreement (see `yardstick::kap()`).
`fun_catg_error`	a function that will evaluate errors for categorical variables (i.e., factors with >2 levels). Default is to use kappa agreement (see `yardstick::kap()`).

a tibble::tibble() with columns variable, type, and score. The score column comprises output from the error functions.

Kappa agreement is a similar to measuring classification accuracy, but is normalized by the accuracy that would be expected by chance alone and is very useful when one or more classes have large frequency distributions.

df_complete <- data.frame(a = 1:10, b = 1:10, c = 1:10, d=1:10,
  fctr = letters[c(1,1,1,1,1,2,2,2,2,2)])

df_miss = df_complete
df_miss[1:3, 1] <- NA
df_miss[2:4, 2] <- NA
df_miss[3:5, 3] <- NA
df_miss[4:6, 5] <- NA


imputes <- list(a=1:3, b=2:4, c=3:5, fctr = factor(c('a','a','b')))

df_imputed <- fill_na(df_miss, vals = imputes)

scored <- scrimp_vars(df_imputed, df_miss, df_complete)