A portable model publishing method, device and storage medium

Abstract

The present invention provides a portable model release method, device and storage medium, the operator stage is acquired, and the operator training model is received, the operator training model includes a plurality of operator units, and the operator training The multiple operator units in the model are divided, and each operator unit is independently encapsulated into an operator stage; the logical relationship between the operator units is obtained, and the logical relationship between the operator units is extracted; the operator release model is established, and multiple In the operator stage and the logical relationship between the operator units corresponding to the operator stage, the operator release model is established; the model file is transformed, and the operator release model is converted into a model file in Json format. The method of the present invention can directly complete the release of the whole process. After the model file of the whole process is converted, only the data needs to be input into the model file, and the data result after the whole process can be obtained, and the operation of the whole process is unnecessary. Need to use a variety of different operating environments to improve the convenience of operation.

Description

technical field [0001] The present invention relates to the field of machine learning, in particular to a portable model publishing method, device and storage medium. Background technique [0002] With the continuous updating of related technologies such as the Internet and big data, people's ability to apply artificial intelligence and other technologies to traditional industries is also rapidly improving. From the initial data mining technology to machine learning to the current hot deep learning technology, the emergence of each technology is impacting traditional technology, prompting the continuous development and updating of technology and business. New technologies are constantly being introduced, and new technology frameworks are also following closely behind. In this era of rapid technological change, we are required to fully consider the adaptability and compatibility of new technologies and new frameworks in the process of product design and project landing design...

AI Technical Summary

Problems solved by technology

[0003] For the technical fields such as model training and model application involved in modern machine learning and deep learning, implementation frameworks and components have also appeared in the industry, such as: Spark MLlib with many machine learning applications, TensorFlow and Pytorch for deep learning, etc., different The core programming languages ​​used by the programming frameworks are also different, some are suitable for Python, some are suitable for java, scala, etc.; different frameworks have their own independent model release methods, and the model training and release technologies between the frameworks are not compatible with each other. As a result, the utilization rate of the model will decrease. The appearance of the new model improves the capability of the model. However, based on the new framework, the downstream business application system may not be able to quickly and smoothly abandon the original model utilization technology and iterate to the new technology platform. In terms of technology, it is difficult to release, which creates barriers to the application of new technologies, and there are obstacles in business promotion

Images