Saivedant HavaandGitHub
dc3eb72b06
The Optuna suggestion service mapped only UNIFORM and LOG_UNIFORM for int and double parameters. A parameter declaring NORMAL or LOG_NORMAL fell through every branch in _get_optuna_search_space and was left out of the search space, so Optuna never sampled it. The hyperparameter was silently dropped from tuning with no error or log. Optuna has no native normal or log-normal distribution, so the backend cannot honor those values. Add an explicit branch for int and double that raises a ValueError naming the parameter and distribution, turning a silent correctness bug into an actionable error. The hyperopt suggestion service already supports these distributions, so the proto values are valid input that this backend must reject explicitly rather than ignore. Adds a parametrized regression test covering NORMAL and LOG_NORMAL for both int and double parameters. Fixes #2665 Signed-off-by: Saivedant Hava <[email protected]>
Kubeflow Katib
Kubeflow Katib is a Kubernetes-native project for automated machine learning (AutoML). Katib supports Hyperparameter Tuning, Early Stopping and Neural Architecture Search.
Katib is the project which is agnostic to machine learning (ML) frameworks. It can tune hyperparameters of applications written in any language of the users’ choice and natively supports many ML frameworks, such as TensorFlow, PyTorch, XGBoost, and others.
Katib can perform training jobs using any Kubernetes Custom Resources with out of the box support for Kubeflow Training Operator, Argo Workflows, Tekton Pipelines and many more.
Katib stands for secretary in Arabic.
Search Algorithms
Katib supports several search algorithms. Follow the Kubeflow documentation to know more about each algorithm and check the this guide to implement your custom algorithm.
| Hyperparameter Tuning | Neural Architecture Search | Early Stopping |
| Random Search | ENAS | Median Stop |
| Grid Search | DARTS | |
| Bayesian Optimization | ||
| TPE | ||
| Multivariate TPE | ||
| CMA-ES | ||
| Sobol's Quasirandom Sequence | ||
| HyperBand | ||
| Population Based Training |
To perform the above algorithms Katib supports the following frameworks:
Prerequisites
Please check the official Kubeflow documentation for prerequisites to install Katib.
Installation
Please follow the Kubeflow Katib guide for the detailed instructions on how to install Katib.
Installing the Control Plane
Run the following command to install the latest stable release of Katib control plane:
kubectl apply -k "github.com/kubeflow/katib.git/manifests/v1beta1/installs/katib-standalone?ref=v0.17.0"
Run the following command to install the latest changes of Katib control plane:
kubectl apply -k "github.com/kubeflow/katib.git/manifests/v1beta1/installs/katib-standalone?ref=master"
For the Katib Experiments check the complete examples list.
Installing the Python SDK
Katib implements a Python SDK to simplify creation of hyperparameter tuning jobs for Data Scientists.
Run the following command to install the latest stable release of Katib SDK:
pip install -U kubeflow-katib
Getting Started
Please refer to the getting started guide to quickly create your first hyperparameter tuning Experiment using the Python SDK.
Community
The following links provide information on how to get involved in the community:
- Attend the bi-weekly AutoML and Training Working Group community meeting.
- Join our
#kubeflow-katibSlack channel. - Check out who is using Katib and presentations about Katib project.
Contributing
Please refer to the CONTRIBUTING guide.
Citation
If you use Katib in a scientific publication, we would appreciate citations to the following paper:
A Scalable and Cloud-Native Hyperparameter Tuning System, George et al., arXiv:2006.02085, 2020.
Bibtex entry:
@misc{george2020katib,
title={A Scalable and Cloud-Native Hyperparameter Tuning System},
author={Johnu George and Ce Gao and Richard Liu and Hou Gang Liu and Yuan Tang and Ramdoot Pydipaty and Amit Kumar Saha},
year={2020},
eprint={2006.02085},
archivePrefix={arXiv},
primaryClass={cs.DC}
}