AutoKeras

Overview

AutoKeras is a free, open-source AutoML package for Python that searches for machine-learning models. It is best suited to developers and data practitioners who want to automate parts of model design while staying in code. Its central trade-off is clear: flexible model search without a software fee, but no graphical workflow.

Key features

The package covers image and text classification and regression, as well as classification and regression for structured data. That makes it a practical option for teams working across several common supervised-learning tasks. Documentation also describes multi-task and multimodal datasets, though using those workflows still means working with the package rather than a visual interface.

AutoKeras lets users build models from its blocks and automate the search for detailed configurations. In image workflows, architecture type, normalization and augmentation can be specified, or left for the system to tune. This is a useful middle ground between a rigid preset and fully manual model configuration, but it assumes familiarity with Python and machine-learning concepts.

The best model found can be exported as a Keras Model. Validation behavior is configurable: by default, the last 20% of training data is reserved for validation, and validation loss guides model selection. Users can choose a different split, supply validation data, and set custom metrics or objectives. For interrupted work, setting overwrite to false lets a run with the same code continue from the contents of its directory.

Pricing

AutoKeras is free: the package costs nothing and is installed with pip. There is no paid tier in the package plan, so the choice is whether a self-hosted Python workflow fits your needs, not which subscription level to buy.

The project provides pip installation instructions. Its installation guide supports Python 3 and recommends PyTorch version 2.3.0 or later when using the PyTorch backend. Those requirements make it a better match for users already prepared to manage a Python environment than for people seeking a managed, no-code setup.

Platforms

AutoKeras supports Linux, macOS and Windows, and is self-hosted. Its workflow interface is code, so users run the package in their own environment rather than through a web-based application.

Who it's for

Choose AutoKeras if you want no-cost automated model search for image, text or structured-data tasks and want to retain control over model components, validation and selection objectives. It is less suitable if you need a graphical workflow or do not want to work directly with Python. The project is developed by DATA Lab at Texas A&M University in collaboration with the keras-team for version 1.0 and above. Questions can be directed to GitHub Discussions.

Pros and cons

• Pros: Free and open source, with image, text and structured-data tasks supported in one Python package.
• Pros: Users can combine AutoKeras blocks with automated configuration search, then export the selected model as a Keras Model.
• Pros: Validation data and model-selection criteria can be customized, and runs can resume from existing directory contents.
• Cons: The code-first, self-hosted workflow is a poor fit for users who want a web interface or managed service.
• Cons: Working with the package requires a Python environment and familiarity with machine-learning workflows.

Alternatives

AutoML Software is a useful starting point if you want to compare a broader set of tools.

FEDOT, LightAutoML, MLJAR AutoML, AutoGluon and Auto-PyTorch are also free AutoML options. LightAutoML and MLJAR AutoML additionally offer web platforms, while FEDOT, AutoGluon and Auto-PyTorch share AutoKeras's Linux and self-hosted availability; Auto-PyTorch is listed for Linux only. These alternatives may suit you better if their platform mix or project licensing is a closer fit.

Amazon SageMaker Autopilot is a paid service with a free trial and API and web platforms, so consider it if you prefer those access modes to a self-hosted Python package. BigML and JADBio are freemium alternatives with free plans and trials; their listed plans include usage or project limits, making them worth considering if you want a service with defined free-tier boundaries.

Verdict

AutoKeras is a strong choice for Python users who want free, configurable AutoML across image, text and structured data. The ability to guide model construction and evaluation without giving up automated search is its main advantage. Look elsewhere if a web interface or managed workflow matters more than self-hosted code.