基于余弦函数局部特征的时间衰变模式

数据流具有无限增长的特征，目前的计算系统无法在线处理整个数据集，只能在有限空间内对部分数据进行处理。为了能够得到尽可能合理的结果，数据流系统常常采用单调递减函数由数据的时间戳来确定数据的权值，根据权值选择数据。广泛使用的单调函数是指数函数和多项式函数，但它们存在衰变速度太快或太慢等问题。提出一种新的时间衰变模式——使用余弦函数的局部衰变速度介于指数和多项式之间的特征来确定数据的权值。实验结果显示相对于指数和多项式衰变，局部余弦衰变具有衰变速度合理、参数易于确定、适用于乱序数据流等优势。

Time decay mode based on degressive cosine ramp

Unlimited growth is one of the main characteristics of data stream.Current computing systems can only process a portion of data instead of the full data set online because of the limited memory and space.In order to obtain reasonable results,decay functions are often used in data stream systems to map the weights of data from timestamps.Monotonic decay functions such as exponential and polynomial functions are widely used,but they decay too fast or too slowly.In this paper,a new decay mode based on cosine function whose decay speed is between exponential function and polynomial function was proposed.The experimental results show that compared to exponential and polynomial decay modes,the degressive cosine ramp decays more reasonably and it is easy to appoint the parameter but also applicable to out-of-order data stream.