The Design And Implementation Of Backend Of Management System For Machine Learning Model

Machine learning algorithms are widely used in data mining,computer vision,speech recognition and other aspects.Due to the differences of computer hardware configuration,the timeliness during machine learning model training process,the differences of machine learning algorithm framework and the complexity of model environment configuration,the traditional machine learning algorithm model generally has the problems of low development efficiency and inadequate utilization of hardware resources.At the same time,due to the high security requirements of the enterprise for the input data and the resulting data of model training,the data transmission and algorithm development process can only be carried out in the internal system.Enterprise internal algorithm engineers need appropriate internal machine learning model management system,which is used to train machine learning model rapidly,judgment of model effect and applied to production,which is of great significance for improving algorithm engineers' development efficiency and saving enterprise resources.This thesis describes the background of the project,summarizes the research and application status of machine learning model management system both at home and abroad,including Spring Boot,Docker,Kubernetes,Ceph,machine learning framework and other related technologies and products involved in this thesis.This thesis analyzes the system requirements in detail,and extracts the functions of user Authentication,namespace management,project management,data storage,model management,model inference and so on,then the system architecture and functional composition are designed.On the basis of private cloud,the system deployment design is carried out,the ER relationship is analyzed in detail,and the system database is designed.In this thesis,the class diagram,sequence diagram and other methods are used to design each functional module in detail,and the implementation details are given.The system described in this thesis has been online,deployed in development environment and production environment respectively,and the system runs stably.The system simplifies the development process and improves the development efficiency.