Deployment and monitoring device and method for machine learning model

Abstract

The invention relates to a deployment and monitoring device of a machine learning model. The deployment and monitoring device comprises: one or more container application modules, wherein a WEB application service interface, a machine learning model code and a python running environment are deployed in each container application module; the container application modules are used for deploying a running environment of a python application program into a container application, providing web interface service for a user side through a flash-based web application framework, and providing web access service for the user side in a Restful API mode; a database module which is configured to receive and store log output in the running process of the machine learning model, wherein the container application module is pre-configured with a running environment of a Python application program, so that the efficiency of the system is improved, and the resource utilization rate is increased.

Description

technical field [0001] The present invention relates to the technical field of machine learning model service deployment, in particular to a device and method for deploying and monitoring machine learning models. Background technique [0002] With the development of cutting-edge hot spots such as big data and artificial intelligence, as well as the accumulation of business data, how to quickly use machine learning and other analysis and mining technologies to discover the value of data assets and let data decisions guide business operations has become a common consideration in the industry and needs to be solved urgently. To solve these problems, major enterprises are actively exploring the application of machine learning and data mining in business operation improvement. [0003] At present, there is no shortage of technical tools for data analysis and mining and machine learning in the industry. Third-party tool libraries for analysis and mining abound (such as scikit-lear...

AI Technical Summary

Problems solved by technology

[0003] At present, there is no shortage of technical tools for data analysis and mining and machine learning in the industry. Third-party tool libraries for analysis and mining abound (such as scikit-learn, TensorFlow, Pytorch, etc.), and the methodological knowledge is relatively complete. These technical tools It often focuses on the feature engineering of data and the realization of algorithm modules. With the help of these tools or methodologies, data cleaning, feature representation, etc. can be quickly realized, but these are only the steps of experimental pre-research or idea verification, and have not formed a user-side Application landing plan to the server

[0004] At the same time, during the running process of the model application, the entire model is like a black box, and it is difficult to monitor the running status and performance of the model service, which is not conducive to the reliability evaluation of the model service, which shows the limitation of the model service in terms of monitorability

The traditional machine learning model application deployment goes through three stages, including: stage 1, install the toolkit and support environment on a physical machine or a virtual machine to form a development environment, stage 2, develop machine learning model code in the development environment, and test , Phase 3. To deploy to the online system after the test is correct, it is necessary to prepare the corresponding toolkit and support environment for the online system. During the whole process, the cost of environment deployment is very high

Images