cheatsheet

Scikit-learn Cheatsheet: Multiclass

Multiclass classification refers to those classification tasks that have more than two targets.

What can be done?

• Convert Binary to Multiclass: Use meta-estimators to apply binary classifiers (like SVM) to multiclass problems.
• Decision Strategy: Choose between “One-vs-Rest” or “One-vs-One”.

Key Algorithms

1. OneVsRestClassifier (OvR):
   • Fits one classifier per class.
   • Computational burden is $N$ classifiers.
   • Most common and scalable strategy.
2. OneVsOneClassifier (OvO):
   • Fits one classifier for every pair of classes.
   • Computational burden is $N \cdot (N-1) / 2$ classifiers.
   • Often used for algorithms that don’t scale well with the volume of data (like kernels).
3. OutputCodeClassifier:
   • Uses Error-Correcting Output Codes. Each class is represented by a binary code.

Native Multiclass support

Some estimators handle multiclass natively (no wrapper needed):

• LogisticRegression (using 'multinomial' loss).
• RandomForestClassifier.
• GaussianNB.
• LinearDiscriminantAnalysis.

Code Snippet: Multiclass Strategies

from sklearn.multiclass import OneVsRestClassifier, OneVsOneClassifier
from sklearn.svm import SVC
from sklearn.datasets import load_iris

X, y = load_iris(return_X_y=True)

# 1. One-vs-Rest
ovr = OneVsRestClassifier(SVC(kernel='linear'))
ovr.fit(X, y)

# 2. One-vs-One
ovo = OneVsOneClassifier(SVC(kernel='linear'))
ovo.fit(X, y)

---

Credits: This cheatsheet is based on the scikit-learn documentation and examples, which are licensed under the BSD 3-Clause License. Copyright (c) 2007 - 2026 The scikit-learn developers. All rights reserved.

This site is open source. Improve this page.