Introduction

SIATune is an open-source deep learning model hyperparameter tuning toolbox especially for OpenMMLab's model frameworks such as mmdetection and mmsegmentation. In order to support job scheduling and resource management, SIATune adopts Ray and Ray.tune.

Major features

• Fully support OpenMMLab models

  We provide a unified model hyperparameter tuning toolbox for the codebases in OpenMMLab. The supported codebases are listed as below, and more will be added in the future.

  • MMPretrain
  • MMDetection
  • MMDetection3D
  • MMRotate
  • MMYOLO
  • MMSegmentation
  • MMOCR
  • MMPose
  • MMHuman3D
  • MMRazor
  • MMFewShot
  • MMAction2
  • MMTracking
  • MMFlow
  • MMEditing
  • MMGeneration

• Support hyperparameter search algorithms

  We provide hyperparameter search algorithms such as below;

  • FLAML
  • HyperOpt
  • Nevergrad
  • Optuna
  • Adaptive Experimentation (AX)
  • Scikit-optimize

• Schedule multiple experiments

  Various scheduling techniques are supported to efficiently manage many experiments.

  • Asynchronous HyperBand
  • HyperBand
  • Median Stopping Rule
  • Population Based Training
  • Population Based Bandits
  • Bayesian Optimization and HyperBand

• Distributed tuning system based on Ray

  Hyperparameter tuning with multi-GPU training or large-scale job scheduling are managed by Ray's distributed compute framework.

Installation and Getting Started

Please refer to get_started.md for installation and getting started.

License

This project is released under the Apache 2.0 license.

Citing SIATune

If you use SIATune in your research, please use the following BibTeX entry.

@misc{na2022siatune,
  author =       {Younghwan Na and Hakjin Lee and Junhwa Song},
  title =        {SIATune},
  howpublished = {\url{https://github.com/SIAnalytics/siatune}},
  year =         {2022}
}