Gini Impurity

noun

say it: “JEE-nee im-PYOOR-ih-tee” /ˈdʒini ɪmˈpjʊrəti/

A measure of how mixed the categories in a group are, from 0 when all are the same to higher values for a more even mix.

What is Gini impurity?

Gini impurity is how mixed a group's classes are: 0 when every example is the same class, 0.5 for an even mix of two. A decision tree picks each question by how much it lowers the impurity of the groups it makes. It's 1 minus the sum of each class's share squared.

def gini(shares):
    return 1 - sum(p ** 2 for p in shares)
print(gini([1.0]))                  # 0.0: all apples
print(gini([0.5, 0.5]))             # 0.5: half and half
print(round(gini([0.8, 0.2]), 2))   # 0.32

In a sentence

  • “The tree chose the question that lowered the Gini impurity the most.”
  • “The split left a group of only apples, with a Gini impurity of 0.”

Level: Freelancer · More AI Basics terms at this level

Updated