基于学习的并行绘制系统中数据集增强算法

Data Set Enhancement Algorithm in Learning-Based Parallel Rendering System

并行绘制系统中,负载平衡是保证其绘制效率的关键因素。为了达到负载平衡,基于随机森林的并行绘制系统在绘制之前,利用训练好的随机森林模型来预测各个子屏的绘制时间,以此为依据找到负载平衡的划分方式。随机森林的预测好坏直接影响寻找平衡划分方式的时间,而想要获得一个预测准确度高的随机森林模型,一个好的训练集是必要因素。因此,基于系统的粗糙采样集,对于每个采样点,进行“邻域”(采样点周围的点)的寻找,再依据“邻域”进行重采样,以此增强训练集。经实验证明,利用该方法对数据集进行增强后,相比于用原始采样方法形成的同样大小的训练集,给随机森林的预测性能带来更大的提升。

In a parallel rendering system,load balancing is the key factor to ensure its rendering efficiency.In order to achieve load balancing,a paral lel forest-based parallel rendering system uses a trained random forest model to predict the drawing time of each sub-screen before draw ing,and uses this as a basis to find a load balancing division method.The quality of the random forest prediction directly affects the time to find a balanced partition.To obtain a random forest model with high prediction accuracy,a good training set is a necessary factor.There fore,the article is based on the system's rough sampling set.For each sampling point,the neighborhood(points around the sampling point)is searched,and then resampling is performed according to the neighborhood to enhance the training set.It has been proved by experiments that after using this method to enhance the data set,compared with the training set of the same size formed by the original sampling method,the prediction performance of the random forest is greatly improved.