Amazon has launched a new toolkit for automated machine learning, AutoGluon, as an open source project designed to make it easier for developers to incorporate deep learning models into their applications.
AutoGluon will serve as an easy-to-use and easy-to-extend auto machine learning (AutoML) with focus on deep learning and real-world applications including text, image, or tabular data. And it is intended for both beginners and experts in machine learning, as it allows quick prototype to deep learning solutions for data with few lines of code.
It leverages on automatic hyper-parameter tuning, model selection / architecture search, and data processing, while automatically utilizing the state-of-the-art deep learning techniques without requiring expert knowledge.
Amazon claims the deployment of deep learning models with state-of-the-art inference, means accuracy will typically require no extensive expertise. As such developers that have had to invest a considerable effort into training deep learning models, now can be rest assured of a ready-made toolset for their application development.
Despite advancements with the Keras library, more easily specifying parameters and layers in deep learning models, developers still have to grapple with complex issues like data pre-processing and hyper-parameter tuning. AutoGluon is intended to fully democratize deep learning and make machine learning more easily available to all developers.
AutoGluon will leverage on available compute resources to locate the strongest model within the allotted runtime. While Python 3.6 or Python 3.7 is required, AutoGluon support is limited to Linux, with MacOS and Windows support still in the works.
It is currently available for tinkering from the project website or GitHub, for developers who want to deep their feet in the waters, even as it guarantees to enable them to produce high-performance neural networking model with as little as three lines of code.