I was wonder how I could reorder the importance of features produced from a trained model from 'mlr3' to match the order of the feature names from task$feature_names
? For example, if I create a task
and model from mlr3
like so:
#Get data
aq <- data.frame(airquality)
aq <- na.omit(aq)
# Create mlr3 task and model
aq_T = TaskRegr$new(id = "aq", backend = aq, target = "Ozone")
aqLrn = lrn("regr.ranger", importance = "permutation")
aq_M <- aqLrn$train(aq_T)
and then I take a look at both the feature names and the importance values, it results in the following:
name <- aq_T$feature_names #task feature names
imp <- aq_M$importance(). #models importance values
name
imp
> name
[1] "Solar.R" "Wind" "Temp" "Month" "Day"
> imp
Temp Wind Solar.R Month Day
597.43488 455.69392 112.31195 30.28683 21.80924
The importance values are ordered by the highest to lowest values. But I was wondering if it's possible to reorder the imp
values to match the order of the feature names given by name
(in the above example).
Taking a look at the structure of both name
and imp
tells me:
str(name)
str(imp)
> str(name)
chr [1:5] "Solar.R" "Wind" "Temp" "Month" "Day"
> str(imp)
Named num [1:5] 597.4 455.7 112.3 30.3 21.8
- attr(*, "names")= chr [1:5] "Temp" "Wind" "Solar.R" "Month" ...
I figured out a way to reorder the named numeric imp
. This did the trick:
imp[order(factor(names(imp), levels = name))]