Label encoding

Categorical data is converted to numeric values with LabelEncoder.

LabelEncoder

Models work with numbers; text labels must be encoded as integers. LabelEncoder gives each distinct value 0, 1, 2... in alphabetical order:

from sklearn.preprocessing import LabelEncoder
le = LabelEncoder()
y = le.fit_transform(["kedi", "köpek", "kedi", "kuş"])
print(y)                 # [0 1 0 2]
print(le.classes_)       # ['kedi' 'köpek' 'kuş']
print(le.inverse_transform([2]))  # ['kuş']

LabelEncoder is meant for labels (y).

One-hot encoding

Unordered categorical features are expanded into a separate 0/1 column for each value:

from sklearn.preprocessing import OneHotEncoder
enc = OneHotEncoder()
X = enc.fit_transform([["kırmızı"], ["mavi"], ["kırmızı"]]).toarray()
print(X)   # [[1. 0.] [0. 1.] [1. 0.]]
print(enc.categories_)

In pandas, pd.get_dummies(df, columns=["renk"]) does the same. For ordered categories (small < medium < large) OrdinalEncoder is suitable.

Exercise Encode the labels

Encode the renk list with LabelEncoder and print the result. Then print the length of classes_.

Show example solution

from sklearn.preprocessing import LabelEncoder

renk = ["yeşil", "mavi", "yeşil", "sarı", "mavi"]

le = LabelEncoder()
print(le.fit_transform(renk))
print(len(le.classes_))

Test yourself

Why can applying LabelEncoder to a feature column (X) be a problem?
- It isn't a problem
- The numbers imply a false order (0 < 1 < 2)
- It is very slow
- It doesn't accept text
Show answer

The numbers imply a false order (0 < 1 < 2) — Linear models treat "kuş" (2) as greater than "kedi" (0); one-hot is used for unordered categories.

Turn the encoded numbers back into text.

adlar = le.____([0, 2])

Show answer

adlar = le.inverse_transform([0, 2])

Match the encoders.

LabelEncoder

Labels to integers
OneHotEncoder

A 0/1 column per value
OrdinalEncoder

Ordered categories
get_dummies

One-hot with pandas

Next lesson: Data splitting

We Are Enhancing Your Experience

Cookie Preferences

Strictly Necessary Cookies

Analytical Cookies