The Study And Implementation Of One-Stop Machine Learning Platform

The development and improvement of machine learning technology(especially deep learning technology),as well as the reduction of the construction cost of large-scale computing clusters and large-scale storage clusters have made machine learning applications in enterprisesmore and more frequent,and the scope of application also becoming more widespread.Although the theoretical foundation of various machine learning algorithms is becoming more and more perfect,and the cost of hardware equipment(CPU,GPU,disks for storing training data,etc.)for applying machine learning is also getting lower and lower,there are still many other problems in the application process of machine learning technologyBuilding a machine learning model is very difficult because machine learning belongs to a relatively professional field.While in Internet companies,the demand is rapidly iterative,for a small demand that needs to use machine learning,also requires the participation of algorithm engineers,to do some algorithm modeling and tuning work.So the labor cost of model development is relatively large.After the completion of model development,it is a great challenge for algorithm engineers who are only proficient in algorithm design and model tuning to make the model working in the project.According to the enterprise machine learning 2020 report from Algorithmia,a well-known Al service provider,many companies face numbers of key issues in deploying machine learning and subsequent service maintenanceAiming at the problems encountered in the practical application of machine learning,this paper makes relevant research and designs a one-stop machine learning platform system.The system provides a one-stop machine learning solution,which can cover the entire process of machine learning business.The core functions of the system include:1.Quickly build a model training job in the form of configuration.2 Provide visual modeling tools,which greatly reduce the threshold of model development.3.A unified model management platform is provided to facilitate algorithmic personnel to manage machine learning models.4.Provide the function of one-key deployment model to online service provision and the monitoring function of model serviceThis platform is mainly developed with JAVA language,and adopts SpringBoot to build the overall project architecture.The back-end database USES relational database MYSQL,non-relational database Tair and so on.The system design conforms to the design idea of high cohesion and low coupling,and the overall code style adopts the official Java code specification standard of alibabaThe platform has been launched in the company and is running smoothly,with increasing users and visits.By today,the platform has 1000+online services,2000+various models and experiments and 100+algorithm components.The launching of the platform has provided a lot of convenience for the algorithm team in the company,greatly reducing the cost of model development and launching.The monthly active users of the platform have reached a considerable number,and the number of new users is also growing steadily,achieving the expected goal of the platform construction.