How to Use OrdinalEncoder in sklearn with Python

Use OrdinalEncoder from sklearn.preprocessing to convert categorical data into ordinal integers. Fit the encoder on your data with fit() or fit_transform(), then transform new data with transform().

Syntax

The basic syntax to use OrdinalEncoder is:

• OrdinalEncoder(): Creates the encoder object.
• fit(X): Learns the categories from data X.
• transform(X): Converts categories in X to ordinal numbers.
• fit_transform(X): Fits and transforms X in one step.

python

from sklearn.preprocessing import OrdinalEncoder

encoder = OrdinalEncoder()
encoder.fit(X)  # Learn categories from X
X_encoded = encoder.transform(X)  # Transform X to ordinal numbers

# Or combine fit and transform
X_encoded = encoder.fit_transform(X)

Example

This example shows how to encode a list of categorical features into ordinal numbers using OrdinalEncoder. It fits the encoder on the data and then transforms it.

python

from sklearn.preprocessing import OrdinalEncoder
import numpy as np

# Sample categorical data with two features
X = np.array([
    ['red', 'S'],
    ['green', 'M'],
    ['blue', 'L'],
    ['green', 'XL']
])

encoder = OrdinalEncoder()
X_encoded = encoder.fit_transform(X)

print("Original data:\n", X)
print("Encoded data:\n", X_encoded)

Output

Original data:
 [['red' 'S']
 ['green' 'M']
 ['blue' 'L']
 ['green' 'XL']]
Encoded data:
 [[2. 0.]
 [1. 1.]
 [0. 2.]
 [1. 3.]]

Common Pitfalls

1. Not fitting before transforming: You must call fit() or fit_transform() before transform(), or you get an error.

2. New categories in transform data: If new categories appear in data passed to transform() that were not seen during fit(), it will raise an error.

3. Data shape: Input must be 2D array-like, even if you have only one feature.

python

from sklearn.preprocessing import OrdinalEncoder
import numpy as np

encoder = OrdinalEncoder()
X_train = np.array([['red'], ['green'], ['blue']])
X_test = np.array([['yellow']])  # New category not seen in training

# Wrong: transform without fit
# encoder.transform(X_train)  # Raises error

# Correct usage
encoder.fit(X_train)

# This will raise error because 'yellow' was not seen during fit
# encoder.transform(X_test)

# To handle unknown categories, use handle_unknown='use_encoded_value' and unknown_value
encoder = OrdinalEncoder(handle_unknown='use_encoded_value', unknown_value=-1)
encoder.fit(X_train)
X_test_encoded = encoder.transform(X_test)
print(X_test_encoded)  # Output: [[-1.]]

Output

[[-1.]]

Quick Reference

Parameter	Description	Default
categories	Specifies categories per feature or 'auto' to infer	'auto'
dtype	Data type of output encoded values	np.float64
handle_unknown	How to handle unknown categories during transform ('error' or 'use_encoded_value')	'error'
unknown_value	Value to use for unknown categories if handle_unknown='use_encoded_value'	None

Key Takeaways

Always fit OrdinalEncoder on training data before transforming.

OrdinalEncoder converts categorical features into integer codes starting at 0.

Handle unknown categories with handle_unknown='use_encoded_value' to avoid errors.

Input data must be 2D array-like, even for single feature columns.

Use fit_transform() to combine fitting and transforming in one step.