FEDOT

Overview

FEDOT is a free, open-source AutoML framework for building data-driven composite models across classification, regression, clustering and time-series forecasting. It suits developers and researchers who want automated pipeline building but still need control over a code-based workflow. Its breadth across preprocessing, optimization and model serialization is compelling; it is less suited to readers seeking a graphical workflow.

Key features

FEDOT covers much of the machine-learning lifecycle: preprocessing, model selection, tuning, cross-validation and serialization. It accepts inputs from CSV files, pandas DataFrames, NumPy arrays and time-series CSV data, and works with tabular, text and image data, including multimodal combinations. Preprocessing handles missing and infinite values, categorical features and extra spaces in categorical data, reducing some routine cleanup before modeling.

Automation is adjustable rather than all-or-nothing. Omit parameters for full automation, or supply them to guide partial automation and manually compose a pipeline. Presets include best_quality, fast_train, stable, auto, gpu, ts and automl; auto is the default. This flexibility is useful when a quick baseline is only the starting point, though making effective use of the controls assumes comfort with code.

Pipeline optimization and learning use GOLEM, which applies meta-heuristic methods to graph-based pipelines. FEDOT draws models mostly from scikit-learn, statsmodels and Keras, and supports integrating libraries including CatBoost and XGBoost, as well as custom libraries. Feature engineering, automated model selection and explainability are included among its capabilities.

The default cross-validation setting uses five folds. Users can add metrics to the optimizer to address potential bias, a useful option when the default evaluation may not suit a particular objective. GPU evaluation uses RAPIDS, but its supported models are limited to Ridge, Lasso, LogisticRegression, RandomForestClassifier, RandomForestRegressor, KMeans and SVC. The GPU preset therefore does not mean every available model can use GPU evaluation.

Install with pip install fedot; optional image, text-processing and DNN dependencies are available through fedot[extra]. The API can also be called from a console without writing Python code, with predictions saved as CSV files. The framework is published under the BSD 3-Clause license, and the maintainers welcome bug reports and enhancement proposals through GitHub issues.

Pricing

FEDOT is free: its plan costs 0.00 USD per free and includes the open-source AutoML framework under the BSD 3-Clause license. There is no paid tier or free-trial period to weigh against a subscription, and no seat or usage caps are given. The trade-off is that users work with the framework themselves rather than choosing a commercial plan with a graphical workflow.

Platforms

FEDOT supports API use on Linux, macOS and Windows, as well as self-hosted deployment. The workflow interface is code-oriented; console invocation provides a route to run the API without Python code, but the output is a CSV rather than a graphical pipeline interface.

Who it's for

FEDOT is a strong fit for developers and research teams who need a free, self-hosted framework for several supervised learning tasks and forecasting, and value the option to switch between automated and parameter-guided pipeline construction. Its multimodal inputs and library extensibility make it relevant to projects that do not fit a single tabular-data workflow. Readers who want a visual interface, or GPU evaluation across a wider model selection, should look elsewhere.

Pros and cons

• Pro: One framework covers classification, regression, clustering and univariate or multivariate forecasting, with tabular, image and text inputs.
• Pro: Users can move from full automation to manually guided pipelines, and GOLEM optimizes graph-based pipeline structures.
• Pro: Free open-source use, self-hosting and integration with established ML libraries provide room to adapt it to project needs.
• Con: The workflow is code-based, which makes it a weaker fit for readers who need a visual interface.
• Con: GPU evaluation covers a defined set of models rather than the framework's entire model range.

Alternatives

EvalML is another free option with API, Linux, macOS, self-hosted and Windows support; consider it if you are comparing code-oriented frameworks on those platforms.

LightAutoML is a free, open-source Python library installable from PyPI under Apache License 2.0, and adds web support alongside Linux, macOS, self-hosted and Windows. Choose it if that license or platform mix better fits your requirements.

FLAML is another free option for API, Linux, macOS, self-hosted and Windows use; it is worth comparing if you want an alternative within that platform set.

AutoGluon is a free open-source Python library under Apache 2.0, supporting Linux, macOS, self-hosted and Windows. It is a reasonable alternative if that license is preferable.

BigML offers a web option and a free plan with unlimited tasks and storage, but caps datasets at 16 MB per task, parallel tasks at two and users at one. Pick it if a web workflow matters and those limits suit your work.

Auto-PyTorch is a free, BSD-licensed option for Linux and self-hosted use, developed by the AutoML Groups of the University of Freiburg and Hannover. Consider it if those are your target platforms.

JADBio offers a free Basic plan limited to one seat, three projects, 50 MB uploads, 500 MB storage and one model export, as well as web support. Choose it if those project limits and its web option suit you.

Akkio is a freemium alternative with API, self-hosted and web support; consider it if you want to compare another option in that category.

Browse more options in AutoML Software.

Verdict

Choose FEDOT if you want a free, self-hosted AutoML framework that spans preprocessing through serialization and lets you tune how much pipeline construction is automated. Its combination of task coverage and graph-based optimization is the central reason to choose it. Look elsewhere if your priority is a visual workflow or broader GPU model coverage.