broken.glm: Breaking Down of Model Predictions for glm models

In pbiecek/breakDown: Model Agnostic Explainers for Individual Predictions

Breaking Down of Model Predictions for glm models

Description

Breaking Down of Model Predictions for glm models

Usage

## S3 method for class 'glm'
broken(
  model,
  new_observation,
  ...,
  baseline = 0,
  predict.function = stats::predict.glm
)

Arguments

model	a glm model
new_observation	a new observation with columns that corresponds to variables used in the model
...	other parameters
baseline	the origin/baseline for the breakDown plots, where the rectangles start. It may be a number or a character "Intercept". In the latter case the orgin will be set to model intercept.
predict.function	function that will calculate predictions out of model (typically predict or betas)

Value

an object of the broken class

Examples

# example for wine data
wine$qualityb <- factor(wine$quality > 5.5, labels = c("bad", "good"))
modelg <- glm(qualityb~fixed.acidity + volatile.acidity + citric.acid +
              residual.sugar + chlorides + free.sulfur.dioxide +
              total.sulfur.dioxide + density + pH + sulphates + alcohol,
    data=wine, family = "binomial")
new_observation <- wine[1,]
br <- broken(modelg, new_observation)
logit <- function(x) exp(x)/(1+exp(x))
plot(br, logit)

# example for HR_data
model <- glm(left~., data = HR_data, family = "binomial")
explain_1 <- broken(model, HR_data[1,])
explain_1
plot(explain_1)
plot(explain_1, trans = function(x) exp(x)/(1+exp(x)))

explain_2 <- broken(model, HR_data[1,], predict.function = betas)
explain_2
plot(explain_2, trans = function(x) exp(x)/(1+exp(x)))

pbiecek/breakDown documentation built on March 15, 2024, 10:46 a.m.