histogram plot from plot.model_performance() is in reverse order

[llendway]

The code below uses the example from the plot.model_performance() code. When used with multiple models and geom="histogram", the labels are incorrect (in reverse order, it seems).

library(patchwork)
library(DALEX)

titanic_ranger_model2 <- ranger(survived~gender + fare, data = titanic_imputed,
num.trees = 50, probability = TRUE)
explainer_ranger2 <- explain(titanic_ranger_model2, data = titanic_imputed[,-8],
y = titanic_imputed$survived,
label = "ranger2")
mp_ranger2 <- model_performance(explainer_ranger2)

titanic_glm_model <- glm(survived~., data = titanic_imputed, family = "binomial")
explainer_glm <- explain(titanic_glm_model, data = titanic_imputed[,-8],
y = titanic_imputed$survived, label = "glm",
predict_function = function(m,x) predict.glm(m,x,type = "response"))
mp_glm <- model_performance(explainer_glm)

titanic_lm_model <- lm(survived~., data = titanic_imputed)
explainer_lm <- explain(titanic_lm_model, data = titanic_imputed[,-8],
y = titanic_imputed$survived, label = "lm")
mp_lm <- model_performance(explainer_lm)

The histogram labels are reversed - the bottom graph should be ranger2 and the top glm.

plot(mp_glm, mp_ranger2, mp_lm, geom = "histogram") +
plot(mp_glm, mp_ranger2, mp_lm, geom = "boxplot")

This is clear after looking at the graph of only ranger2.

plot(mp_ranger2, geom = "histogram") +
plot(mp_ranger2, geom = "boxplot")

[hbaniecki]

Indeed, this is due to this line:

DALEX/R/plot_model_performance.R

Lines 141 to 142 in 922519c

	# if factor, then levels shall be reversed
	if (length(levels(df$label)) > 1) levels(df$label) <- rev(levels(df$label))

@pbiecek any idea why it is there? I found a mismatch of colours in EMA too. Can we delete it?

[pbiecek]

thanks,
in fact labels were not consistent
and it was a problem not only in model_performance() but also model_parts()

it should be fixed in 24ba9cf

library(patchwork)
library(DALEX)
library(ranger)

titanic_ranger_model2 <- ranger(survived~gender + fare, data = titanic_imputed,
                                num.trees = 50, probability = TRUE)
explainer_ranger2 <- explain(titanic_ranger_model2, data = titanic_imputed[,-8],
                             y = titanic_imputed$survived,
                             label = "ranger2")
mp_ranger2 <- model_performance(explainer_ranger2)

titanic_glm_model <- glm(survived~., data = titanic_imputed, family = "binomial")
explainer_glm <- explain(titanic_glm_model, data = titanic_imputed[,-8],
                         y = titanic_imputed$survived, label = "glm",
                         predict_function = function(m,x) predict.glm(m,x,type = "response"))
mp_glm <- model_performance(explainer_glm)

titanic_lm_model <- lm(survived~., data = titanic_imputed)
explainer_lm <- explain(titanic_lm_model, data = titanic_imputed[,-8],
                        y = titanic_imputed$survived, label = "lm")
mp_lm <- model_performance(explainer_lm)
mp_pa <- model_parts(explainer_lm)
mp_pa2 <- model_parts(explainer_glm)
mp_pa3 <- model_parts(explainer_ranger2)

plot(mp_glm, mp_ranger2, mp_lm, geom = "histogram") +
  plot(mp_glm, mp_ranger2, mp_lm, geom = "boxplot") +
  plot(mp_pa2, mp_pa3, mp_pa)

[pbiecek]

fixed in 24ba9cf