應用自動化測試於異質環境機器學習管道之 MLOps 系統;An MLOps system that applies automated testing on ML pipelines in heterogeneous environments.

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/93390

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/93390

Title:	應用自動化測試於異質環境機器學習管道之 MLOps 系統;An MLOps system that applies automated testing on ML pipelines in heterogeneous environments.
Authors:	黃宗泓;Huang, Zong-Hong
Contributors:	資訊工程學系
Keywords:	MLOps;機器學習部署;機器學習管道;管道測試;MLOps;ML Deployment;ML pipeline;Pipeline test
Date:	2023-07-31
Issue Date:	2024-09-19 16:57:04 (UTC+8)
Publisher:	國立中央大學
Abstract:	在將機器學習方案應用於實際環境時，我們面臨著各種挑戰。 MLOps（Machine Learning Operations）的實踐建議可以幫助維運人員快速將機器學習方案部署到生產環境中。但事實上，現有的MLOps 平台功能仍不完善。在建置管道的過程中，需要將每個機器學習模組打包成容器，而無法容器化的模組存在使用限制，並且現有平台也未提供測試功能來檢測管道的正確性，因此需要進行人工端到端測試。為了解決這些問題，我們提出了一個新的MLOps 平台。該平台能夠在異質環境上建立管道元件，從而讓更多的機器學習方法可以透過平台建置成管道；此外，我們的平台還提供不同等級的自動化測試功能，以測試機器學習管道的正確性。本文將通過比較現有平台，來闡述我們平台在加速機器學習管道部署方面的優勢。同時，我們將通過一個實際的機器學習部署案例來說明，我們平台提供的功能在該案例的部署過程中所帶來的效益。;When applying ML(Machine Learning) solutions in production environments, we face various challenges. The recommendations of MLOps (Machine Learning Operations) can assist operators in rapidly deploying ML solutions to production. However, the existing MLOps platforms is still incomplete. In the process of building pipelines, it is necessary to containerize each ML module, and modules that can’t be containerized have usage limitations. Additionally, the current platforms don’t provide testing functionality to verify the correctness of pipelines, thus requiring manual end-to-end testing. We propose a new MLOps platform to address these issues. This platform enables more ML methods to be built as pipelines in heterogeneous environments. Furthermore, our platform offers automated testing functionality at different levels to test ML pipelines. In this paper, we will illustrate the advantages of our platform in accelerating the deployment of ML pipelines by comparing it with existing platforms. Additionally, we will demonstrate the benefits of our platform during the deployment process through a practical case study.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	19	View/Open

社群 sharing

Loading...