Application Of Moran Index And Deep Q Network In Classifier Model Approximation And Simulation

On the Internet,organizations will open their own technology and application for the use of potential customers online on the site by means of fees or restricted free,but for the sake of cost and stability,it's needed to create their own services,so the model approximation and simulate demand arises at the historic momentu which uses intelligent algorithm to the black box model,eventually emersion the function of the model.In recent years,deep reinforcement learning(DRL)has been applied to games,vedios and achieved great achievements.It can be seen that DRL has a good performance in the field of machine vision.For example,Taking the picture classifier open on baidu AI platform,this paper proposes a proxy IP invocation strategy based on Moran's I,which accelerates the interactive learning process of agents based on deep Q network(DQN),and finally completes the approximation and simulation of this classifier.First,this paper finishing collected from the internet free agent to release the website of the proxy IP as IP pool,use a more general foreign IP as an agent,in the form of proxy crawler visit AI platform open interface,solves the problem of limited domestic agent in a part of the site visit and AI platform for IP speed limit access problems.Secondly,a nearby call strategy based on Moran's I is proposed,which makes spatial statistics on the geographic location and delay of proxy IP.The local Moran's I and test numbers of China are 0.98 and 2.21,which proves that the response speed of proxy IP has positive spatial autocorrelation in the surrounding areas of China.After 30 interaction exper-iments,the consumption of nearby strategy is 57 seconds less than that of random strategy without continuous testing,which solves the problem of excessive delay and server load in traditional strategy,alleviates the problem of excessive training cycle of agents.Finally,DQN agent based on proxy IP is used to interact with AI platform.Compared with traditional deep learning,DQN reinforcement learning mechanism has advantages of being more general and stable.After 13288 interactions,the similarity effect reached 85.5%on the test set,and the simulation model of AI platform classifier was obtained,which proved the validity of the model approximation and simulation in this paper.