[PaddlePaddle Hackathon] 94 Implement AutoDL for PaddlePaddle based on KubeFlow Katib

[tizhou86]

（This is a task ISSUE for PaddlePaddle Hackathon, for more details, see PaddlePaddle Hackathon）

【Task Specification】

• Title：Implement AutoDL for PaddlePaddle based on KubeFlow Katib

• Tag：Framework，C++，Python，Hyper-Parameter Tuning

• Difficulty：Medium

• Detailed Description： Katib is an inborn project of Kubernetes designed for Automatic Deep Learning (AutoDL). It supports hyper-parameter tuning and neural architecture search. Currently, Katib supports deep learning frameworks, including TensorFlow, MXNet, PyTorch, XGBoost, etc. For this project, PaddlePaddle AutoDL based on Kubeflow Katib should be implemented.

  Steps：

  1. Download and compile the source code of Katib from GitHub.
  2. Understand content of the task by referring to sample codes.
  3. Read related documents and codes to understand how to implement PaddlePaddle AutoDL by leveraging Katib.
  4. Complete unit testing and related demo.
  5. Generate PR and related documents.

【Content Submission】

• Project PR to Katib
• Related technical documents

【Technical Requirements】

• Proficiency in C++，Python
• Understanding of hyper-parameter tuning in deep learning

[gaocegege]

I think this proposal is to support Paddle in Katib, it is helpful for both paddle and katib community 🥂

[andreyvelich]

Will be very nice to see integration between Paddle and Katib.
Thank you for driving this @tizhou86!

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[stale]

This issue has been automatically closed because it has not had recent activity. Please comment "/reopen" to reopen it.