Research Of Analysis Pattern Driven Parameter Optimization

Enormous commercial value brought by big data analysis promotes the vigorous development of big data analysis industry.However,due to the lag in data analysts'awareness of big data and the lack of analytical experience,it has severely restricted the effectiveness and accuracy of big data analysis.To help inexperienced data analysts accurately and completely define the problem to be analyzed,reuse existing analytical experience,and efficiently find the best path from the problem domain to the solution domain.This paper combines the problem framework and the goal-based requirements modeling technology.An analysis pattern driven analytical requirements modeling method is proposed.Parameter optimization is one of the key steps to achieve intelligent data analysis.In order to simplify the parameter optimization analysis process,this paper proposes an analysis pattern driven parameter optimization method.Firstly,the analysis pattern driven analytical requirements modeling method is used to analyze the relevant requirements of parameter optimization,and the parameter optimization Analysis goal-Operationalizing Soft-goal Dependency Diagram is constructed to form the goal satisfiability metric;secondly,the parameter optimization analysis pattern is introduced to reuse the empirical data of parameter optimization;Thirdly,under the guidance of the parameter optimization analysis pattern,according to the satisfiability metrics of the parameter optimization analysis goals,Precisely locating and reusing existing parameter optimization models in the parameter optimization model repository,Reducing excessive error attempts by data analysts in the selection of parameter optimization algorithms and implementations,which improves the efficiency and quality of parameter optimization;Finally,custom parameter optimization model parameters and run parameter optimization.In addition,we improve the existing modeling tool Star-UML,establishes an analysis pattern repository,an analysis model repository,a metrics repository,and realizes the weaving function of the analysis pattern.With the help of this improved analysis tool,the analysis pattern driven parameter optimization method is applied to the parameter optimization process of image classification model training.Comparing with the optimal parameter solution set obtained by the brute force calculation,it is shown that the optimal parameter solution set calculated by the parameter optimization model chosen in this paper is almost the same as that obtained by the brute force calculation,and the time used is significantly shorter.Therefore,it can be confirmed that applying the method proposed in this paper can quickly find an effective parameter optimization solution,thus illustrating the application value of this method.