l*****0 发帖数: 299 | 1 如何同时fit两条相近的lines?
数据模型 y=px+a+rand, y=qx+b+rand, rand遵循高斯分布(zero mean,sigma相同)
p,q,a,b,sigma未知
两条线斜率p,q相近(但又都 >0),两线在data points附近部不交叉
data points的noise较大,很多夹在两线中间的数据点不知是属于哪条线的。
请问高如何做fitting?
谢谢! | B****n 发帖数: 11290 | 2 I am not sure I totally understand your question. It seems that you need to
find a change point between these two lines and estimate the coefficients p,
q, a,b.
If this is what you want, there are bunch of methods dealing with it. For
example, you can use the segmented regression package to do it; below is the
link for the segmented package using free software R.
http://cran.r-project.org/web/packages/segmented/segmented.pdf
【在 l*****0 的大作中提到】![](/moin_static193/solenoid/img/up.png) : 如何同时fit两条相近的lines? : 数据模型 y=px+a+rand, y=qx+b+rand, rand遵循高斯分布(zero mean,sigma相同) : p,q,a,b,sigma未知 : 两条线斜率p,q相近(但又都 >0),两线在data points附近部不交叉 : data points的noise较大,很多夹在两线中间的数据点不知是属于哪条线的。 : 请问高如何做fitting? : 谢谢!
| N******t 发帖数: 11 | 3 Robust statistics
Hough transform | l*****0 发帖数: 299 | 4 非常谢谢您的答复!大概是我没把问题说清。不是segmented regression。
例如:
数据由两个隐藏模型同时产生y1=2x+1+rand1, y2=2.1x+1.1+rand2,
其中rand1和rand2的mean为0,sigma相同但大小未知。
所有数据位于0
的,如何做fitting?
谢谢。
to
p,
the
【在 B****n 的大作中提到】![](/moin_static193/solenoid/img/up.png) : I am not sure I totally understand your question. It seems that you need to : find a change point between these two lines and estimate the coefficients p, : q, a,b. : If this is what you want, there are bunch of methods dealing with it. For : example, you can use the segmented regression package to do it; below is the : link for the segmented package using free software R. : http://cran.r-project.org/web/packages/segmented/segmented.pdf
| l*****0 发帖数: 299 | 5 Thanks a lot for your reply. 数据的噪音较大Hough transform does not work
well.
【在 N******t 的大作中提到】![](/moin_static193/solenoid/img/up.png) : Robust statistics : Hough transform
| B****n 发帖数: 11290 | 6 This is a mixture normal model and the parameters can be estimated using
EM algorithm.
The following video introduces the mixture model (very similar to yours) and
how to estimate its parameters.
http://www.youtube.com/watch?v=LBtuYU-HfUg
【在 l*****0 的大作中提到】![](/moin_static193/solenoid/img/up.png) : 非常谢谢您的答复!大概是我没把问题说清。不是segmented regression。 : 例如: : 数据由两个隐藏模型同时产生y1=2x+1+rand1, y2=2.1x+1.1+rand2, : 其中rand1和rand2的mean为0,sigma相同但大小未知。 : 所有数据位于0: 的,如何做fitting? : 谢谢。 : : to : p,
| N******t 发帖数: 11 | 7 How much noise? More than 50%?
http://arxiv.org/abs/1208.1810 | f*****e 发帖数: 2992 | 8 PRML neural networks那一章, mixture density networks,那个可真是优美。
【在 l*****0 的大作中提到】![](/moin_static193/solenoid/img/up.png) : 如何同时fit两条相近的lines? : 数据模型 y=px+a+rand, y=qx+b+rand, rand遵循高斯分布(zero mean,sigma相同) : p,q,a,b,sigma未知 : 两条线斜率p,q相近(但又都 >0),两线在data points附近部不交叉 : data points的noise较大,很多夹在两线中间的数据点不知是属于哪条线的。 : 请问高如何做fitting? : 谢谢!
|
|