Code-generated machine learning model full-process automatic deployment method and system

Abstract

The invention discloses a code-generated machine learning model full-process automatic deployment method and system. The method comprises the steps of obtaining full-process step information of a first machine learning model; obtaining a data processing operator of a single step; obtaining first online software, and obtaining a first online software environment; further generating a first character string template; extracting the trained parameters in the data processing operator to be input into the first character string template for rendering, and generating a single-step code text; generating a first deployment structure; constructing a first general prediction template; and obtaining a first deployment result based on the single-step code text. The technical problem that the real-time prediction performance of the machine learning model cannot be further improved due to the fact that automatic deployment of the whole process of the machine learning model cannot be achieved based on the code generation technology in the prior art is solved. The technical effect of realizing automatic deployment of the whole process of the machine learning model by utilizing a code generation technology in a compiling principle is achieved.

Description

technical field [0001] The present application relates to the field of computers, and in particular, to a method and system for automatic full-process deployment of a code-generated machine learning model. Background technique [0002] In production, algorithm engineers often need to train models and deploy the trained models to online software systems, that is, to convert the calculation results of some three-party open source frameworks or internal frameworks of enterprises on data into an online service, which can Get the same effect as the offline model. However, machine learning models running in production environments are complex software systems. Unlike ordinary software development and deployment, machine learning engineers face some new challenges. For example, complex model pipelines are composed of different data processing operations, which usually contain many parameters; even multiple model pipelines are integrated, which seriously aggravates the difficulty ...

AI Technical Summary

Problems solved by technology

[0005] The purpose of this application is to provide a code-generated machine learning model full-process automatic deployment method and system to solve the problem in the prior art that the automatic deployment of the whole process of the machine learning model cannot be realized based on the code generation technology, so that the machine learning model cannot be further improved. Technical Issues of Learning Models for Real-time Predictive Performance

Images