Skip to content

Combine_likelihood_scans.md

Latest commit

 

History

History
75 lines (57 loc) · 3.53 KB

Combine_likelihood_scans.md

File metadata and controls

75 lines (57 loc) · 3.53 KB
Raw

Likelihood scan with freezing

  1. First of all perform a initial fit and save the snapshot

combine -M MultiDimFit combined.root -n .snapshot --rMin 0 --rMax 2 --saveWorkspace -t -1 --expectSignal=1 --robustFit=1 --toysFreq

  1. Perform a likelihood scan without freezing any parameter

combine -M MultiDimFit combined.root -n .scan --rMin 0 --rMax 2 --saveWorkspace -t -1 --expectSignal=1 --robustFit=1 --toysFreq --algo grid --points 30

  1. Freeze all nuisances and perform the scan loading best fit values from snapshot

combine -M MultiDimFit combined.snapshot.root -n .freezeAll --rMin 0 --rMax 2 -t -1 --expectSignal=1 --robustFit=1 --toysFreq --algo grid --points 30 --freezeParameters "rgx{.*}" --snapshotName MultiDimFit

  1. Plot the two plots:

python $COMBINE_TOOLS/plot1DScan.py initial_scan.root --others 'freezed_scan.root:FreezeAll:2' -o freeze_second_attempt --breakdown Syst,Stat

2D likelihood scan

The following commad performs a scan on the signal strength and a nuisance parameters with a MC asimov dataset

combine   -d workspace.root  \ 
-M MultiDimFit  \
--redefineSignalPOIs QCDscale_wjets,r   \
--setParameterRanges QCDscale_wjets=-3,3:r=0.5,1.5  \
 -t -1 --expectSignal=1 \
--algo grid -m 125 \
--points 1000

The combineTool can be used to speed up the process

combineTool.py -M MultiDimFit  -d workspace.root \
 --redefineSignalPOIs QCDscale_wjets,r  \
 --setParameterRanges r=0,2:QCDscale_wjets=-3,3 \
-t -1 --expectSignal=1  -n "mu_QCDscaleWjets" \
-m 125 --algo grid --points 1000 \
--job-mod condor --task-name qcdscale_mu --split-point 10

Scan on Likelihood saving all the nuisances profiling mu

Perform a likelihood scan on a parameter but save all the values of the profiles nuisances

combineTool.py -d workspace.root  -M MultiDimFit    \
 --algo=grid   --X-rtd OPTIMIZE_BOUNDS=0  --rMin 0.5 --rMax 1.5 \
--saveSpecifiedNuis all   -n "name"   \
--points 400    --job-mode condor --task-name task-name --split-points 1 
--trackParameters "rgx{.*_norm_.*}"

be careful to include -t -1 --setParameters r=1 if you want to fit the asimov dataset and not the data.

In order to track also the rate params another option needs to be added with a regex

--trackParameters "rgx{.*_norm_.*}"

To plot the 2D points use the script plot2Dscan_params:

Scan on Likelihood profiling one nuisance parameter

Perform scan on NUisance parameter saving the value of all the others nuisances and r. Useful to check the correlations. Using the --redefineSignalPOI all the constraint terms on the nuisance are removed and not considered in the fits. If the constraints need to be included use the option --parameters nuisance to select the POI for the scan.

combineTool.py -d workspace.root -M MultiDimFit  --algo=grid \
 --redefineSignalPOI  NUISANCE --setParameterRanges r=-2,3:NUISANCE=-1.2,1.2\
 --setParameters r=1  -n "PSscan" --points N --saveSpecifiedNuis all\
 --trackParameters "rgx{.*},r"  -t -1 \
--job-mode condor --task-name task-name --split-points 1

The --setParameters r=1 is needed to build the asimov dataset