Container-based model training test tuning and deployment method and device

Abstract

The present invention provides a container-based model training test tuning and deployment method and device, the method comprising the following steps: deploying a service grid service to the container environment where the model training is located; receiving the interface service through the model training and testing algorithm driver Request input parameters for algorithm iteration; in the process of model training and test algorithm iteration, record the correlation between algorithm iteration parameters and training objective functions, and record and visualize the intermediate results in the training process. The method of the present invention improves the security by using the container service grid conversion method, and the request of the algorithm is served by the request interface, and there will be no problem that it cannot be embedded in the original system. At the same time, the parameters of the model training and the training process The corresponding relationship of the training objective function is analyzed and stored, which is convenient for reverse tuning of the model parameters according to the performance indicators of the model. In addition, the intermediate process of the model training is visualized, which is convenient for debugging and training optimization.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular to a container-based model training test optimization and deployment method and a container-based model training test optimization and deployment device. Background technique [0002] Currently, the container network matching and conversion method provides a convenient and flexible calling method for driving algorithm training and testing programs through interface services outside the container. The current container algorithm debugging is mainly performed by jupyter notebook remote calling. The security and the compatibility of the original system are unsatisfactory. In addition, the current correspondence between algorithm training parameters and model evaluation capabilities mainly depends on the developer's own records, and it is not easy to trace the parameter adjustment and optimization process of model training. Contents of the invention [0003] In order to sol...

AI Technical Summary

Problems solved by technology

[0003] In order to solve the above technical problems, the present invention provides a container-based model training test tuning and deployment method, which can improve the security by using the container service grid conversion method, and the request of the algorithm is served by the request interface, without There will be a problem that it cannot be embedded in the original system. At the same time, the corresponding relationship between the parameters of model training and the training objective function in the training process is analyzed and stored, which is convenient for reverse tuning of model parameters according to the performance indicators of the model. In addition, the intermediate process of model training Visual display for easy debugging and training optimization

Images