Examples with TUnfold

[fzeiser]

Is it possible to create a toy model (example) where one can also test TUnfold?

On the current master, 8977e40, I fixed #1 to get TUnfold working. However, running the sample data I would receive following output:

./toyModel/testBiasTUnfold.py nobias && mv *.eps results_TUnfold_regFirstDer/
This software has been produced using the ideas put forth in:
Choudalakis, G., ``Fully Bayesian Unfolding'', physics.data-an:1201.4612, https://arxiv.org/abs/1201.4612
Please cite it if you plan to publish this.
More information on the software itself can be found in https://github.com/daniloefl/Unfolder

Checking bias due to configuration 'A'
Info in <TUnfold::SetConstraint>: fConstraint=0
Info in <TUnfold::TUnfold>: underflow and overflow bin do not depend on the input data
Info in <TUnfold::TUnfold>: 4 input bins and 4 output bins
TUnfold::TUnfold:0: RuntimeWarning: too few (ny=4) input bins for nx=4 output bins
Info in <TUnfoldDensity::TUnfold>: *NOT* unfolding bin #0 (yaxis:[ufl])
Info in <TUnfoldDensity::TUnfold>: *NOT* unfolding bin #5 (yaxis:[ofl])
Info in <TUnfoldDensity::RegularizeOneDistribution>: regularizing yaxis regMode=2 densityMode=0 axisSteering=*[UOB]
TUnfoldDensity::SetInput:0: RuntimeWarning: 4/4 input bins have zero error, and are ignored.
TUnfoldDensity::SetInput:0: RuntimeWarning: no data to constrain output bin #1 (yaxis:[-0.5,0.5])
TUnfoldDensity::SetInput:0: RuntimeWarning: no data to constrain output bin #2 (yaxis:[0.5,1.5])
TUnfoldDensity::SetInput:0: RuntimeWarning: no data to constrain output bin #3 (yaxis:[1.5,2.5])
TUnfoldDensity::SetInput:0: RuntimeWarning: no data to constrain output bin #4 (yaxis:[2.5,3.5])
Error in <TUnfoldDensity::SetInput>: 4/4 output bins are not constrained by any data.
scanRegParameter: parameter =  0.0  /  0.19
getBiasFromToys: Throwing toy experiment 0/1000
TUnfoldDensity::GetInputInverseEmatrix:0: RuntimeWarning: number of parameters 4 = input rank 4. Problem is ill posed
TUnfoldDensity::GetInputInverseEmatrix:0: RuntimeWarning: number of parameters 4 = input rank 4. Problem is ill posed
TUnfoldDensity::GetInputInverseEmatrix:0: RuntimeWarning: number of parameters 4 = input rank 4. Problem is ill posed
[...]

[daniloefl]

Hi @fzeiser

I had not seen your message before. I have changed the toy model a few times to do other tests and I didn't check the effect in TUnfold. I believe that the problem here is that TUnfold expects that the number of reco bins is at least twice larger than the number of truth bins.

The FBU method does not impose this constraint, so when I changed the model to look into the FBU results, I did not think of the effect it would have in TUnfold. As I am not a TUnfold expert, I am not sure to which extend (or under which configurations) this requirement is imposed. This is actually a reason why it is hard to make this comparison.

You can change toyModel/ModelChrisSmallVar2.json to have a migration matrix that is compatible with 2 times as many reco bins and also change the input data in the lines below.

Best regards,
Danilo

  "Data" : [335, 290, 195, 100]

    , "Mig" :
      [ [0.80, 0.10, 0.00, 0.00]
      , [0.05, 0.80, 0.10, 0.00]
      , [0.00, 0.05, 0.80, 0.10]
      , [0.00, 0.00, 0.05, 0.80]
      ]