Double Machine Learning Model Training Optimization Research Based On The Serverless Architecture

With the widespread application of machine learning methods in various fields,the double machine learning model for causal modeling of data has been widely used.It has played an important role in the fields of statistics,econometrics,and COVID-19 forecast-ing.DML model allow researchers to estimate and reason about causal parameters by exploiting the superior predictive power of machine learning methods in an efficient sta-tistical framework.From the perspective of model use,the DML model needs to use two machine learning methods to debias the data,but in the actual use of the model,there is no accurate algorithm to select the best one for different data to be predicted.Machine learning methods,which lead to large deviations in the prediction results of the model.From the perspective of model deployment and training,the traditional cloud computing model,due to its own resource usage and service cost limitations,makes it difficult for the DML model to effectively exert the predictive performance of the model in some high-parallel scenarios.Therefore,how to effectively select the best machine learning method to improve the prediction accuracy of the DML model,and at the same time effectively reduce the deployment and training cost of the DML model in the cloud computing mode and improve the model prediction performance is an important challenge for this paper.In order to solve the problems raised above,this paper designs and implements a DML model training system based on serverless architecture.The specific contributions are as follows:(1)Through Monte Carlo simulation experiments,the prediction accuracy of five com-mon machine learning methods in the DML model is compared,and a dynamic machine learning selection algorithm based on sample characteristics is proposed,which effectively improves the prediction accuracy of the DML model.(2)By combining serverless architecture with DML model deployment and training,it effectively solves the problems of poor prediction performance and high service cost of DML models under the traditional cloud computing model.The character-istics of serverless computing,such as elastic scaling and On-demand payment,can effectively reduce the cost of using the model while improving the service perfor-mance of the DML model.(3)Based on the Alibaba Cloud serverless computing platform,a general,flexible,and high-performance DML model training system λ-DML is designed and imple-mented,so that the DML model can efficiently exert the predictive ability of the model under the serverless architecture.In order to verify the advantages of the algorithm and system proposed in this paper in improving the prediction accuracy of the DML model,service performance,and reduc-ing the cost of use,this paper conducts a series of comparative experiments by deploying the DML model in the traditional Iaa S architecture and the serverless architecture.The experimental results show that the dynamic function selection algorithm and λ-DML sys-tem proposed in this paper can effectively improve the prediction accuracy and service performance of the model while reducing the cost of using the DML model.