AutoKeras

AutoKeras is an AutoML library for deep learning, built on Keras and TensorFlow by the Keras team. With more than 9,300 stars on GitHub, it automates the hardest parts of building neural networks, searching architectures and hyperparameters so users can go from raw data to a trained model in a few lines of code.

Overview

AutoKeras treats model design as a search problem. You declare the shape of your inputs and the task, whether image classification, text classification, regression or structured data prediction, and the system explores network architectures and training settings using efficient neural architecture search guided by network morphism. When the search finishes, you get a normal Keras model that you can inspect, fine-tune and export for serving. Pre-built blocks for common building patterns keep searches fast, and the API stays close enough to everyday Keras that experienced users never feel locked out of the details.

Key Features

• Automated neural architecture search with efficient network-morphism-based exploration
• Ready-made AutoModel pipelines for image, text and structured/tabular tasks
• Composable search blocks for custom architectures and multi-input, multi-output models
• Automatic preprocessing, including text vectorization and feature handling
• Exports a standard Keras model you can deploy with TensorFlow Serving or TFLite
• Officially maintained by the Keras team with documentation and tutorials

💳 Pricing

AutoKeras is free and open source under the Apache License 2.0. There is no paid version; your only costs are the compute used during architecture search and training.

Who Is It For

AutoKeras is for data scientists who need strong baseline models quickly, and for developers without deep architecture expertise who still want custom neural networks rather than off-the-shelf APIs. It is also a practical teaching tool for understanding how architecture choices affect model quality.