The Design And Implementation Of Model And Variable Analysis System Based On Django

With the wide application of models in the machine learning area, the overall quality of models has largely exerted impact on the ultimate accuracy of information mining. However, it is common to find there are gaps between the design phase and implementation phase of modeling, and differences between various data sources. These elements may possibly cause some kind of risks in the quality aspect of the final model. So it is quite essential to emphasize the audit of model variables and output scores in the modeling process. At present, the traditional audit process requires users to have the detailed knowledge of the whole execution and there are complicated command steps in the background console. These issues have caused this audit process hard to maintain and reuse.This model and variable analysis system based on Django is intended to solve the above problems. This system’s basic function is to make sure the accuracy of models (in terms of model specs) and variables (in terms of variable values). Meanwhile, it also provides analysis reports about variables’ distribution condition through several data analysis methods. There are three basic functional modules in this system, including PMML Simulation, Variable Track, and Distribution Analysis.This system uses Django framework to develop the front end and meets all the functional requirements through model, template and view sections. As for the back end data analysis part, it uses shell scripts and some data analysis libs of python, such as pandas.So overall, this system can help users have more understanding of the quality as for the risk models and variables. What’s more, it has greatly simplified the previous audit process through the standardized interfaces and steps, as well as the easy-to-use web portal. It offers the users an effective channel to submit tasks and receive final analysis reports. Also, it can be applied to other situations where comprehensive data analysis is needed.This thesis firstly introduces the overall background of this project, including the topic source and the meaning of this research. Then it gives brief introduction of some related technologies used by this system, such as Django framework, some data analysis libs of python and shell scripts. What’s more, it makes requirement analysis for the whole system by summarizing its functional and non-functional requirements and using the use case diagram to do further illustration. In the design part, it firstly gives the architectural design by the overall structural diagram, flow chart, and activity diagram. Then it makes detailed design for each module, including PMML Simulation module、Variable Track module. Two-way Variable Comparison module, and One-way Variable Distribution module. It elaborates the functional design by the core class diagram and the collaborative design by the sequence diagram. Finally, it describes the detailed implementation for each module, including the code for some pivotal parts and the screenshots for some key functions.