c*u
发帖数: 916 | 1 我有一组实验数据，
y = y_high, if x > thr
y = y_low, if x < thr
y_high, y_low, thr 是未知的常数，thr是所需要得到的东西．好像3D bruteforce
search 寻找最低点不是很有效率．
举个通俗的例子，假设x是温度，y是水的体积，那么Thr是熔点．但是实验数据中
y的测量有很大noise, 所以必须拟合．
谢谢， | s***t
发帖数: 195 | 2 maybe statistical classification is a better method.
【在 c*u 的大作中提到】
: 我有一组实验数据，
: y = y_high, if x > thr
: y = y_low, if x < thr
: y_high, y_low, thr 是未知的常数，thr是所需要得到的东西．好像3D bruteforce
: search 寻找最低点不是很有效率．
: 举个通俗的例子，假设x是温度，y是水的体积，那么Thr是熔点．但是实验数据中
: y的测量有很大noise, 所以必须拟合．
: 谢谢，
| c*u
发帖数: 916 | 3 Which statistical method do you mean?
I just did a try. It looks like iteration works pretty well. Just need make
a guess of y_high and y_low, find the best thr. Then set y_high, y_low as
the mean values for xthr, find the thr again. Repeat the
iterations until converge. Only takes 2-5 iterations to converge.
Calculation
time is O(n) instead of O(n^3). For my data, it seems the fitting rarely
stuck in local maxima.
Sorry for the dumb question.
【在 s***t 的大作中提到】
: maybe statistical classification is a better method.
|
|
