NNI is an all-inclusive AutoML toolkit designed to automate the machine learning lifecycle. It includes feature engineering, model compression, neural architecture search, and hyper-parameter tuning.

This powerful yet lightweight toolkit helps manage AutoML experiments and dispatches trial jobs generated by tuning algorithms for finding the best neural architecture. It also finds hyper-parameters in several training environments like Remote Servers, Local Machine, Kubeflow, OpenPAI, DLWorkspace, AdaptDL, and many other cloud options.

NNI serves as a great toolkit for those truing different AutoML algorithms in training models and those running AutoML trial jobs to accelerate the search.

Data scientists and researchers can use the toolkit to implement new AutoML algorithms, such as neural architect search algorithm and hyperparameter tuning algorithm.

Project Background

• • • Library: NNI AutoML Toolkit
    • Author: Microsoft
    • Initial Release: 2017
    • Type:  Manages AutoML experiments
    • License: MIT License
    • Contains: Supported libraries and frameworks, algorithms
    • Language: Python
    • GitHub: /microsoft/nni
    • Runs On: Windows, macOS, Ubuntu
    • GitHub Discussions: /microsoft/nni/discussions
    • Stackflow: /questions/8792607/getting-started-with-neural-networks-ann

Applications

• Hyperparameter Tuning
•  General NAS Framework
• Model Compression
• Automatic Feature Engineering
• Performance measurement, comparison and analysis