Add plot.model_profile for two or more models

[hbaniecki]

plot.model_profile in DALEX/ingredients uses the color parameter to choose a dimension for plot splitting, e.g. by the model (label). This allows for comparing two models in one plot.

library(DALEX)

titanic_glm_model <- glm(survived~., data = titanic_imputed, family = "binomial")
explainer_glm <- explain(titanic_glm_model, data = titanic_imputed, verbose = FALSE)
expl_glm <- model_profile(explainer_glm)

library("ranger")
titanic_ranger_model <- ranger(survived~., data = titanic_imputed, num.trees = 50,
                               probability = TRUE)
explainer_ranger  <- explain(titanic_ranger_model, data = titanic_imputed, verbose = FALSE)
expl_ranger <- model_profile(explainer_ranger)

plot(expl_ranger$agr_profiles, expl_glm$agr_profiles, color = "_label_") 
plot(expl_ranger$agr_profiles, expl_glm$agr_profiles)

It would be nice to have similar functionality in survex

ve_rf <- model_profile(explainer_rf, variables = "age")
plot(ve_rf)
ve_cox <- model_profile(explainer_cox, variables = "age")
plot(ve_cox, ve_rf) # won't work

Firstly, at this moment the above code should return an error

The idea is to add some facet parameter, which by default is set to variables but can be changed to _label_ etc.

plot(ve_cox, ve_rf, facet="_label_")

Mockup:

Using:

(plot(ve_rf) + ggtitle("rsf")) +
  (plot(ve_cox) + ggtitle("cph")) + 
  plot_annotation(subtitle="")

[mikolajsp]

I agree this is useful functionality. I'm just wondering how to show this information in one plot.

Do you think one column per model would be a good way to show it? and then different variables in rows?

[hbaniecki]

Perhaps, the most popular use cases consider 1-2-4 variables and 1-2-3 models.