并行绘制系统中基于随机森林的预测算法改进  

An Improved Prediction Algorithm Based on Random Forest in Parallel Rendering System

作　　者：郭赛赛 李君怡 GUO Sai-sai;LI Jun-yi(College of Computer Science,Sichuan University,Chengdu 610065)

机构地区：[1]四川大学计算机学院,成都610065

出　　处：《现代计算机》2020年第6期27-31,共5页Modern Computer

摘　　要：并行绘制系统中,负载平衡是影响绘制效率的关键因素。系统运行时的负载通过随机森林(RF)进行预测,并根据预测结果调整子任务划分方式,实现负载平衡,提高绘制帧率。预测结果越准确,负载调度过程越快速。随机森林的预测结果由组成它的决策树中的部分叶子节点均值决定。随机森林中的决策树在构建过程中不考虑过拟合的问题,因此得到的随机森林的准确性也会存在一定偏差。为了解决这个问题,将部分中间节点加入最终结果的预测中去,通过拟合度判定函数去决定一棵决策树用来预测的节点。实验证明,该方法可以提升随机森林的预测准确率。In parallel rendering system,load balancing is the key factor that affects rendering efficiency.The load of the system running is predicted through the Random Forest(RF),and the subtask division mode is adjusted according to the predicted results to achieve load balancing and improve the rendering frame rate.The more accurate the prediction,the faster the load scheduling process.The prediction result of a ran dom forest is determined by the mean value of some leaf nodes in the decision tree that makes up the random forest.In the process of con structing the decision tree in the random forest,the problem of overfitting is not considered,so the accuracy of the obtained random forest will have a certain deviation.In order to solve this problem,some intermediate nodes are added into the prediction of the final result,and the predicted node of a decision tree is determined by the overfitting decision function.Experimental results show that this method can im prove the prediction accuracy of random forest.

关 键 词：并行绘制系统 负载平衡 随机森林 决策树 过拟合 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]