Method, device, and service server for replacing operation after model update

Abstract

Disclosed are a method, device and service server for replacing and running a model after updating, the method comprising: comparing the comprehensive performance of each updated model; determining the updated model with the best comprehensive performance as a candidate model; Comparing the performance of the candidate model with the model currently running online at a specified risk threshold, wherein the performance of the candidate model or the model currently running online at the specified risk threshold is based on its response to a specified online event The standard score corresponding to the score output by the set, and the audit results of the specified online event set are determined according to the standard score and the specified risk threshold, and the specified risk threshold is determined by the preset model strategy; if compared If it is found that the performance of the candidate model is better than that of the model currently running online, the candidate model is deployed as a model running online.

Description

technical field [0001] The embodiments of this description relate to the technical field of data processing, and in particular to a method, device, and service server for replacing and running a model after it is updated. Background technique [0002] The essence of intelligent or automatic processing is the process in which the computer obtains the corresponding output information after calculation according to the given input information. The model is a kind of information used to describe the above-mentioned input-output relationship. A simple model can be based on Artificial rules are established. In the data age, data modeling based on big data and machine learning technology has become the mainstream. At present, in various application fields, big data modeling has been widely used to realize various intelligent processing functions. With the automatic update of the model and the replacement of the old and new models, it is also necessary to re-deploy the model policy...

AI Technical Summary

Problems solved by technology

However, since the independent model is trained based on certain historical data samples, as time goes by, there may be various factors that lead to the decline of the applicability of the independent model, so it is necessary to update the independent model in a timely manner. Correspondingly, for the integration The model also needs to be updated in a timely manner. For example, a refit update is performed on each integrated model (including the currently officially running integrated model) every week. After each integrated model is updated, it is necessary to re-select an integrated model with the best performance. Replace the currently officially running integrated model and redeploy the model strategy at the same time. However, at present, more manual intervention is still required in the evaluation and selection of the model and the deployment of the model strategy in the model replacement operation stage, resulting in a longer time-consuming model replacement operation cycle. , which results in the updated integration model not being able to function in a timely manner

Images