bottom_up_segmentation

  这是TD算法的逆过程，首先将时间序列，划分为相邻点的短序列，当然此时的拟合误差为0，因为第一点和第二点的连线，原始点都落在线段上。将相邻两个线段连接起来，此时每条线段包含三个原始点，计算中间那个点的拟合误差。这样，所有这些三个点的线段中的中间点的拟合误差计算出来后，找出误差最小且误差小于阀值R的分段，作为第一条包含三个点的线段。 在上面的基础上，第一条分段同样的和相邻线段连接，然后计算每一条分段的拟合误差，再找出误差最小且小于阀值R的分段，作为第二个分段。依次方式循环，直到所有分段的拟合误差都小于阀值R，分段结束。 (This is the reverse process of TD algorithm, the first time series is divided into short sequences adjacent dots, of course fitting error is zero at this time, because the first and second points of connection, the original point fall line on. The two adjacent segments are connected, then each segment contains three original point, the intermediate fitting error calculated that point. Thus, after all these segments fitting error of three points in the intermediate point calculated to find the minimum error and the error is less than the threshold R segment, as the first line contains three points. Based on the above, the first segment and the adjacent line segment connecting the same, and then calculate a fitting error for each segment, and then find the smallest error and less than the threshold R segment, as the second segment . Followed by a cyclic manner until all segments fitting error is less than the threshold R, segment ends.)