Advanced usage of TLinearFitter?

[Georg_T]

Hi,
I’m using TLinearFitter in order to fit datasets in a 3-dim space. I’m using a polynom up to grade 2.

I have a few questions:

1. I’m trying to illustrate the results by fixing two variables of the fit and make several TF1 out of the 3-dim fit. Is there a common method to do this. Currently, I’m calculating the results manually by using GetParameters(). What I’m searching for is

TF1 * TLinearFitter:GetTF1(int fixed par, Double_t p1, Double_t p2, Double_t p3,…) // The entry of the fixed parameter is regardless

2. Why isn’t there an Double_t TLinearFitter::Evaluate (Double_t p1, Double_t p2, Double_t p3, …) Function, as it is known from TF1??

3. Is there an easy method to access the confidence level of the fit (in dependency of the parameters? I would like to make a 1d representation (see question 1) including the error limits

4. How do I access the R^2 (Representing the fit-quality). Usually R^2 (always smaller than 1, but close to 1 means good fit)

Thanks
Georg

---

Please read tips for efficient and successful posting and posting code

_ROOT Version:6.22
Platform: Windows 10
_Compiler: VS xx

---

[eguiraud]

Hi @Georg_T,
and sorry for the late reply!

We need @moneta’s help on this one, let’s ping him.

Cheers,
Enrico

[moneta]

@Georg_T

1. What I’m searching for is

TF1 * TLinearFitter:GetTF1(int fixed par, Double_t p1, Double_t p2, Double_t p3,…)

I guess you would like a profiled function, i.e. a TF1 that is a function of the fixed parameter and the TLinearFitter finds the minimum value for the other two.
There is not such function available, but one can implement this, for example using a lambda or a functor that deals with create the lines fritter, fixing the parameter value and find the solution.

@Georg_T

Why isn’t there an Double_t TLinearFitter::Evaluate (Double_t p1, Double_t p2, Double_t p3, …) Function, as it is known from TF1??

Why this should evaluate ? To the chisquare function given the parameter values ?
For this you can call TLinearFitter::GetChisquare().

@Georg_T

Is there an easy method to access the confidence level of the fit (in dependency of the parameters? I would like to make a 1d representation (see question 1) including the error limits

No, you can compute confidence intervals on the parameters as is done in Minos (see point 1) or otherwise why not running Migrad+Minos after the linear fitter ?

@Georg_T

How do I access the R^2 (Representing the fit-quality). Usually R^2 (always smaller than 1, but close to 1 means good fit)

I don’t know what is R^2.