tidy up dataframe containing model feature importance

returns dataframe with exactly two columns, vars and imp and aggregates dummy encoded variables. Helper function called by all functions that take an imp parameter. Can be called manually if formula for aggregating dummy encoded variables must be modified.

Usage

tidy_imp(imp, df, .f = max, resp_var = NULL)

Arguments

imp

dataframe or matrix with feature importance information

df

dataframe, modeling training data

.f

window function, Default: max

resp_var

character, prediction variable, can usually be inferred from imp and df. It does not work for all models and needs to be specified in those cases.

Value

dataframe

vars

character column with feature names

imp

numerical column, importance values

Examples

# randomforest
df = mtcars2[, ! names(mtcars2) %in% 'ids' ]
m = randomForest::randomForest( disp ~ ., df)
imp = m$importance
tidy_imp(imp, df)
#> # A tibble: 10 × 2
#>    vars     imp
#>    <chr>  <dbl>
#>  1 cyl   96402.
#>  2 mpg   89923.
#>  3 hp    77633.
#>  4 wt    71970.
#>  5 drat  46440.
#>  6 gear  21548.
#>  7 qsec  20524.
#>  8 vs    15734.
#>  9 carb   8570.
#> 10 am     5091.