无法使用手电筒 package 与 R 中的虹膜数据计算交互？

Question

So Im trying to use the flashlight package with the iris data. One of the functions flashlight provides is to calculate the interactions between variables called light_interaction . I can get light_interaction to work if I use an lm model but I can not get it to work if I create my model using mlr3 [side note: if I create a regression model using mlr3 , it works perfectly with flashlights interaction function... I just can't get it to work for classification... and I need flashlight to work with mlr3 for my package]. In the code below I show an example of it working with an lm model... and an example of it not working with mlr3 ... am I doing something wrong?

lm model example that works:

library(flashlight)

fit <- lm(Sepal.Length ~ ., data = iris)
x <- flashlight(model = fit, label = " ", data = iris, y = "Species")
light_interaction(x, pairwise = TRUE, type = "H", grid_size = 5,
                         normalize = F)

mlr3 model that doesn't work:

library(mlr3)
library(mlr3learners)
# mlr3 TASK
bc_T = TaskClassif$new(id = "dat", backend = iris, target = "Species")
# learner
lrn = lrn("classif.ranger")
# model
bc_M <- lrn$train(bc_T)

x <- flashlight(model = bc_M, label = " ", data = iris, y = "Species")
light_interaction(x, pairwise = TRUE) # this line creates an error

This throws back an error saying:

Error in rowsum.default(xx * ww, gg): 'x' must be numeric In addition: Warning message: In Ops.factor(xx, ww): '*' not meaningful for factors

Answer 1

mlr3 uses R6 and the fitted model is stored in the $model slot. I do not know what flashlight::light_interaction() does but as far as your error is concerned, it is as simple as

library(mlr3)
library(mlr3learners)
library(flashlight)
bc_T <- TaskClassif$new(id = "dat", backend = iris, target = "Species")
lrn <- lrn("classif.ranger")
bc_M <- lrn$train(bc_T)

x <- flashlight(model = bc_M$model, label = " ", data = iris, y = "Species")
print(x)
#> 
#> Flashlight   
#> 
#> Model:            Yes
#> y:            Species
#> w:            No
#> by:           No
#> data dim:         150 5
#> predict_fct default:  TRUE
#> linkinv default:  TRUE
#> metrics:      rmse
#> SHAP:             No

^{Created on 2020-11-12 by the reprex package (v0.3.0)}