Hi,
I’m trying to migrate my C++ fitting script into python and encountered a problem doing eigen-analysis on the covariance matrix.
The following code snippet:
fun = ROOT.TF1("func", "expo", 2.5, 9.5)
fitres = hist_data.Fit("func", "IRS")
# deal with uncertainties:
C = fitres.GetCovarianceMatrix()
C.Print()
EigenC = ROOT.TMatrixDEigen(C)
results in:
FCN=1115.07 FROM MIGRAD STATUS=CONVERGED 54 CALLS 55 TOTAL
EDM=1.06952e-09 STRATEGY= 1 ERROR MATRIX ACCURATE
EXT PARAMETER STEP FIRST
NO. NAME VALUE ERROR SIZE DERIVATIVE
1 Constant 1.37419e+01 9.78097e-03 5.02296e-05 5.11911e-04
2 Slope -9.08180e-01 2.58631e-03 1.32823e-05 1.97096e-02
2x2 matrix is as follows
| 0 | 1 |
0 | 9.567e-05 -2.401e-05
1 | -2.401e-05 6.689e-06
Traceback (most recent call last):
File “test.py”, line 33, in
EigenC = ROOT.TMatrixDEigen(C)
TypeError: none of the 3 overloaded methods succeeded. Full details:
TMatrixDEigen::TMatrixDEigen() =>
takes at most 0 arguments (1 given)
TMatrixDEigen::TMatrixDEigen(const TMatrixT& a) =>
could not convert argument 1
TMatrixDEigen::TMatrixDEigen(const TMatrixDEigen& another) =>
could not convert argument 1
So, I gather the TFitResult object is properly retrieved and I can produce the covariance matrix of the fit. However, TMatrixDEigen constructor does not seem to work with the argument it should accept, to my understanding. The C++ version reads:
…
TMatrixD C = fitres->GetCovarianceMatrix();
TMatrixDEigen* EigenC = new TMatrixDEigen(C);
TMatrixD E = EigenC->GetEigenValues();
TMatrixD V = EigenC->GetEigenVectors();
…
and works for me alright.
I clearly must have missed something here.
Any hints will be much appreciated.
Many thanks and best wishes,
pawel bruckman
Hello,
What’s the type of C in the snippet? Can you print(type(C))?
Also, what ROOT version are you using?
Hi,
Print(type(C) gives:
<class ‘ROOT.TMatrixTSym’>
and I’m currently using ROOT 6.16/00
Thanks!
pawel
Hi, I’m sorry. Perhaps I should have replied to you directly. Anyhow, I replied to your questions in a general post. I’m new to the forum 
Hi,
First I think it would be worth trying it with a newer version of ROOT (>= 6.22).
But looking at the code, it seems that the type of the result of GetCovarianceMatrix is TMatrixTSym<double>, aka TMatrixDSym. However, the constructor of TMatrixDEigen accepts a TMatrixT.
Well, this is true. But I thought TMatrixDSym derived from TMatrixT. Otherwise, it wouldn’t work in C++ Root, would it? Is there a way to force an appropriate type conversion? I shall check it with a higher Root version, as you recommend.
p.
According to:
https://root.cern.ch/doc/v622/classTMatrixTBase.html
Both TMatrixTSym and TMatrixT are subclasses of TMatrixTBase, but there is no direct inheritance between them. Am I right that TMatrixDSym is TMatrixTSym<double>?
This I do not know for sure, but very likely it is TMatrixTSym. My trouble is that in C++ Root TMatrixDEigen accepts both TMatrixDSym and TMatrixD in the constructor. I just checked it explicitly.
I also checked that in PyRoot ROOT.TMatrixDEigen behaves the same no matter if you pass the TMatrixTSym or TMatrixT. The problem seems in interpreting the reference “&”. This is what I get when TMatrixT is passed to the constructor:
<class ‘ROOT.TMatrixT’>
Traceback (most recent call last):
File “test.py”, line 40, in
EigenC = ROOT.TMatrixDEigen(C)
TypeError: none of the 3 overloaded methods succeeded. Full details:
TMatrixDEigen::TMatrixDEigen() =>
takes at most 0 arguments (1 given)
TMatrixDEigen::TMatrixDEigen(const TMatrixT& a) =>
could not convert argument 1
TMatrixDEigen::TMatrixDEigen(const TMatrixDEigen& another) =>
could not convert argument 1
Maybe there is a simple workaround I’m not aware of?
I temporarily solved my problem by simply converting the TMatrixDSym object into a numpy array. Then I used scipy.linalg.eigh to solve my eigenproblem.
Still, this is not a satisfactory solution to the original issue, of course 
p.
The missing link
TMatrixT(const TMatrixTSym <Element> &another);
Would adding the following line be helpful:
TMatrixD EigenC = ROOT.TMatrixDEigen(C)
or
EigenC = ROOT.TMatrixDEigen(TMatrixD(C))
Hi Eddy,
Bingo! The explicit type conversion does work indeed (just adding the ROOT namesapace):
EigenC = ROOT.TMatrixDEigen(ROOT.TMatrixD(C))
The first option throws a syntax error, though.
Thanks a lot for this very helpful suggestion 
Best,
pawel
first option should have been:
EigenC = TMatrixD(ROOT.TMatrixDEigen(C))
No, this version still has the same problem with ROOT.TMatrixDEigen(C) which is not accepted.
So presumably, we are left with the second option. Still fair enough!
Thanks again,
p.