- Home
- AutoML Software
- AutoKeras
AutoKeras
What it is
AutoKeras is a free, open-source AutoML system based on Keras that lets users build machine-learning models through code. It supports image and text classification and regression, along with classification and regression for structured data. Its documentation also includes a tutorial for multi-task or multimodal datasets. Users can assemble models from AutoKeras blocks and let the system search detailed configurations. In image workflows, they can set architecture, normalization, and augmentation choices or leave those decisions to automatic tuning. The best model found can be exported as a Keras Model. AutoKeras is installed as a Python package with pip and supports Linux, macOS, Windows, and self-hosted use. The installation guide supports Python 3 and recommends PyTorch 2.3.0 or later when using the PyTorch backend. By default, the final 20% of training data is used for validation, though users can choose another split or provide validation data. Model selection uses validation loss by default, with options to set other metrics or objectives. To continue a prior fit, users can run the same code with overwrite disabled.
For and against
For
- Free and open source.
- Supports image, text, and structured data tasks.
- Can export the selected model as a Keras Model.
- Users can customize validation data and objectives.
- Runs on Linux, macOS, and Windows.
Against
- Workflow is code-based.
- PyTorch 2.3.0 or later is recommended for its PyTorch backend.
- Default validation uses the final 20% of training data.
The verdict
AutoKeras provides automated model search across several task types while allowing users to customize blocks and selection criteria. Users should be comfortable with a Python-based workflow and its training configuration options.
Who it suits
AutoKeras suits developers who want code-based tools to automate model selection and configuration for supported data tasks. It also offers an option for users working with multimodal or multi-task datasets.
Plans and prices
Open-source library
Install locally with pip · no hosted-service limits stated
Read 15 Oct 2026 · opencollective.com
AutoKeras
Python package installed with pip
Read 4 Oct 2026 · autokeras.com
On the record
- Community funding
- Open Collective reports $2,065.59 raised and $205.67 disbursed for AutoKeras.opencollective.com · read 15 Oct 2026
- Community size
- Open Collective lists 32 contributors.opencollective.com · read 15 Oct 2026
- Company location
- The Texas A&M Engineering page identifies the institution in College Station, Texas.engineering.tamu.edu · read 15 Oct 2026
- Custom search spaces
- Users can connect built-in blocks or implement their own Block subclasses to customize the search space.autokeras.com · read 15 Oct 2026
- Deployment ecosystem
- The paper states exported models can be deployed through TensorFlow ecosystem tooling.jmlr.org · read 15 Oct 2026
- Documentation depth
- The documentation includes task tutorials, API reference pages, customization guidance, export instructions and an FAQ.autokeras.com · read 15 Oct 2026
- Funding
- Texas A&M reported that the project had been funded by DARPA through an AFRL contract.engineering.tamu.edu · read 15 Oct 2026
- Image classification
- The task API supports image-classification workflows.autokeras.com · read 15 Oct 2026
- Image features
- Image workflows can configure architecture type, normalization, and data augmentation, or leave choices to be tuned automatically.autokeras.com · read 4 Oct 2026
- Image regression
- The task API supports image-regression workflows.autokeras.com · read 15 Oct 2026
- Installation
- The official installation method is pip install autokeras, with Python 3 required.autokeras.com · read 15 Oct 2026
- Latest package release
- PyPI lists version 3.0.0 as released on November 12, 2025.pypi.org · read 15 Oct 2026
- Licence
- The source repository and package metadata specify Apache License 2.0.github.com · read 15 Oct 2026
- Limit: structured-data production fit
- The same Reddit discussion reported relying on boosted trees for tabular data and not finding neural-network success in production.reddit.com · read 15 Oct 2026
- Maintainer
- The package is developed by DATA Lab at Texas A&M University in collaboration with keras-team for version 1.0 and above.autokeras.com · read 4 Oct 2026
- Maker
- The package is developed by DATA Lab at Texas A&M University with Keras-team collaboration.autokeras.com · read 15 Oct 2026
- Model design
- Users can build custom models from AutoKeras blocks while letting the system search for detailed configurations.autokeras.com · read 4 Oct 2026
- Model export
- The best searched model can be exported as a Keras model.autokeras.com · read 15 Oct 2026
