Re: root for regression

From: Rene Brun (Rene.Brun@cern.ch)
Date: Tue Dec 07 1999 - 18:34:16 MET


Hi Christian,
Root is not currently instrumented to achieve what you want
automatically.
You have to give some initial values for the parameters.

We had developped many years ago a very nice program called MUDIFI.
This program was written in Fortran and it would be a nice addition to
Root.
I have no time to do this work now.
However, if somebody is interested, I can provide some guidance.

Rene Brun

Christian Kaiser wrote:
> 
> Hi,
> I have to perform a regression to experimental data, which is presented in a
> number of vectors of dimensionality five (x1,x2,x3,x4,x5,x6). This data should
> be approximated by a function of the form
> 
>  x6=a * x1^b * x2^c * x3^d * x4^e * x5^f  + g*x1^h.... and so on
> 
> with parameters (a,b,c ...). The problem is, that fitting with an algorithm like
> finding the steepest gradient from the starting-point to minimize chi squared,
> is not possible, because I have no guess of appropriate starting points. I would
> be grateful, if someone could tell me, if ROOT was able to deal with this kind
> of problem und if it has advances compared with some standard software like SAS.
> Thanks in advance
> Christian Kaiser



This archive was generated by hypermail 2b29 : Tue Jan 04 2000 - 00:43:44 MET