Version 3.0: Built for Reproducible Science

The Global Lab for Machine Learning Research

Machine learning thrives on transparency. OpenML is the open, collaborative environment where scientists share FAIR data, organize experiments, and build upon state-of-the-art algorithms.

Trusted worldwide to benchmark algorithms objectively.

+99.99% Reproducibility 500k+ Datasets 10M+ Runs+

The Ecosystem

Frictionless Integration

Seamlessly import data and export experiments from your native scientific environment.

View API Documentation

The Three Pillars of OpenML

Open platform for sharing datasets, algorithms, and experiments to build a global machine learning repository

FAIR Data at Scale

Access thousands of uniformly formatted datasets. Every dataset is versioned, meta-tagged, and ready for immediate loading into your analysis pipeline.

Objective Evaluation

Stop guessing. Run your algorithms on standardized tasks with predefined train/test splits to ensure your results are comparable and peer-review ready.

Frictionless ML Flows

Treat experiments as objects. Share your model pipelines (flows) and results (runs) automatically to create a transparent scientific record.

Accessibility & Integration

Use OpenML from the web UI, notebooks, or the command line—wherever you already work.

The Web Interface

Discover OpenML through an interactive web dashboard for exploring datasets and experiments.

Search, filter, and bookmark datasets by task type, size, and domain.
Inspect metadata, target variables, and distributions before you ever write code.
Compare versions, track provenance, and jump directly into related tasks, flows, and runs.

Browse, visualize, and organize everything from one place—ideal for exploratory analysis, teaching, and quick demos.

Diagram illustrating the scientific machine learning workflow and data lifecycle

Browse Datasets

The Code (APIs)

Integrate directly into your code. Use our client libraries to programmatically download datasets, run tasks, and upload results without leaving your IDE.

Python

import openml

# Load the diabetes dataset by ID
dataset = openml.datasets.get_dataset(37)
X, y, categorical_indicator, attribute_names = dataset.get_data(
  target=dataset.default_target_attribute
)

# Print dataset information
print(f"Dataset: {dataset.name}")
print(f"Features: {len(attribute_names)}")
print(f"Instances: {len(X)}")

View on GitHub Download ZIP

View API Docs

Benchmarking SuitesThe Scientific Method for Rigorous Evaluation

Validate your models on curated benchmarking suites spanning domains such as healthcare, finance, and computer vision. Ensure your algorithm is robust across many datasets, not just a single benchmark.

Tasks

The Problem

Standardized machine learning challenges with fixed evaluation metrics (AUC, RMSE, Accuracy). Each task provides predefined train/test splits to ensure your results are comparable and peer-review ready.

Suits (or Suites)

The Collection

Curated collections of tasks (e.g., "AutoML Benchmark," "Medical Diagnosis Suite") for comprehensive algorithm stress-testing across multiple datasets and problem domains.

The OpenML Workflow LoopIntegrate OpenML into every step of your ML workflow.

Import

Load OpenML datasets in a single line of code, compatible with scikit-learn, PyTorch, TensorFlow, XGBoost, and more.

Build & Run

Train and evaluate models on those tasks using your favorite ML libraries. Try different pipelines and hyperparameters until you get results you are happy with.

Export & Publish

Automatically publish: Upload your workflows (flows), experiments (runs), and evaluation metrics back to OpenML so others can compare and reuse them.

→ Flowsworkflows → Runsexperiments → Metricsevaluation

Contribute to the Global Knowledge Base

Generate persistent identifiers (DOIs) for your datasets, workflows, and experiments—ensuring FAIR compliance and enabling reproducible science.

The Citation Lifecycle

Don't let your research die on a hard drive. Uploading to OpenML creates citable, versioned artifacts with provenance tracking. Your work becomes part of the global ML benchmark corpus, cited by peers and powering meta-research.

OpenML FAQ

Clear answers to the most important questions

Whether you're training your first model or running large-scale benchmarks, OpenML streamlines every step of your workflow. This Q&A highlights what makes OpenML unique and how it can simplify your daily ML work.