cheatsheet

Scikit-learn Cheatsheet: Datasets

The datasets module provides utilities to load standard datasets, fetch data from external repositories, and generate synthetic data for benchmarking and testing.

What can be done?

• Quick Prototyping: Use “toy” datasets like Iris or Digits.
• Benchmarking: Fetch large real-world datasets from OpenML.
• Algorithm Testing: Generate synthetic datasets with controlled noise, clusters, or informative features.

Categories of Datasets

1. Toy Datasets: Loaded immediately with the library.
   • load_iris(), load_digits(), load_wine(), load_breast_cancer().
2. Real-world Datasets: Downloaded on demand.
   • fetch_20newsgroups(), fetch_lfw_people() (faces), fetch_california_housing().
3. OpenML: Access thousands of datasets from openml.org.
   • fetch_openml(name='mnist_784').
4. Synthetic Generators:
   • make_classification(): For binary/multiclass problems.
   • make_regression(): For regression problems.
   • make_blobs(): For clustering.
   • make_moons(), make_circles(): For non-linear separation tests.

Tips

• as_frame=True: Many loaders have this argument to return data as a pandas.DataFrame instead of a NumPy array.
• return_X_y=True: Returns (X, y) directly instead of a Bunch object.

Code Snippet: Generating and Loading Data

from sklearn.datasets import make_classification, load_iris, fetch_openml

# 1. Load Toy Dataset as DataFrame
iris = load_iris(as_frame=True)
df = iris.frame

# 2. Generate Synthetic Classification Data
X, y = make_classification(
    n_samples=1000, 
    n_features=20, 
    n_informative=15, 
    n_redundant=5, 
    random_state=42
)

# 3. Fetch from OpenML
# version='active' or specific number
mnist = fetch_openml('mnist_784', version=1, as_frame=False)

---

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.