auto-sklearn
automl.github.io
- Free plan
- Linux
- API
- Self-hosted
What auto-sklearn is
auto-sklearn is free, open-source software that automates model fitting and ensemble construction for classification and regression. Auto-sklearn 2.0 selects models, decides whether to use Successive Halving, and applies portfolios for meta-learning. Fitted pipelines are compatible with the scikit-learn API. Users can customize metrics, train-validation splits, and feature types, inspect the search procedure, and extend the system by registering wrapper classes for classifiers, regressors, and preprocessing methods. Documentation lists pip, conda-forge, source, and Docker installation options. The tool requires Linux and cannot run on Windows because it relies on Python’s Unix-only resource module. The project does not actively support macOS because of resource-limit and dependency issues. Users must set time and memory limits, and one CPU core is used by default. Parallel workers require a shared file system for training data and models. The free plan lists Linux, Python >=3.7, and a C++11 compiler as requirements.
Who it suits
It suits developers and machine-learning practitioners working on Linux classification or regression workflows who want an open-source AutoML tool compatible with scikit-learn. It is not suited to Windows users.
Good to know, and what to watch
- Free open-source software
- Automates classification and regression model fitting
- Scikit-learn-compatible fitted pipelines
- Custom metrics and validation splits
- Multiple documented installation methods
- Cannot run on Windows
- macOS is not actively supported
- Time and memory limits must be set
- Parallel workers need a shared file system
Plans
Every plan and priceFree open-source software
Open source
The verdict
auto-sklearn offers configurable AutoML for classification and regression within the scikit-learn ecosystem. Linux is required, and resource limits plus shared storage requirements matter for larger or parallel runs.
What the maker publishes
- AutoML method
- Auto-sklearn 2.0 automatically selects models, decides whether to use Successive Halving, and uses portfolios for meta-learning. source
- Customization
- Users can customize metrics, train-validation splits, feature types, and inspect the search procedure. source
- Distributed runs
- Parallel auto-sklearn workers require access to a shared file system for training data and models. source
- Extensibility
- Users can add classification, regression, and feature-preprocessing methods by implementing and registering wrapper classes. source
- Install options
- The documentation describes pip, conda-forge, source, and Docker installation options. source
- macOS support
- The project says it does not actively support macOS because of resource-limit and dependency issues. source
- Maker
- The AutoML website identifies AutoML as a project of the Chair for Machine Learning at the University of Freiburg. source
- Resource limits
- Users must set time and memory limits, and auto-sklearn uses one CPU core by default. source
- Scikit-learn compatibility
- Its fitted pipelines are compatible with the scikit-learn API. source
- Security
- The opened product documentation describes time and memory resource limits but does not state security certifications or compliance claims. source
- What it does
- auto-sklearn automates machine learning model fitting and ensemble construction for classification and regression tasks. source
- Windows support
- The documentation says auto-sklearn cannot run on Windows because it relies on Python’s Unix-only resource module. source
Checked against the maker's pages on 9 Oct 2026.