RooFitUtils

This package is a collection of tools for advanced workspace manipulation and fitting using RooFit. Among others, it includes

• ExtendedMinimizer and ExtendedModel, advanced fitting tools originally designed by Stefan Gadatsch
• editws, workspace manipulation functions designed by Tim Adye.
• guessCorrelations, a helper script able to intelligently identify nuisance parameters when combining workspaces.

Setup

This package supports setup within RootCore and CMake based ASG releases as well as standalone compilation with ROOT.

In order to compile with CMake, type

mkdir build
cd build
cmake .. -DCMAKE_MODULE_PATH=$ROOTSYS/cmake/
make -j4
cd ..
source setup.sh

Now, you are ready to use the package. Don't forget to

source setup.sh

every time you create a new shell.

It is recommended to have python 2.7 or higher. scipy and scikit-image packages are required for plotting likelihood results.

First make sure you have pip installed

curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py
python get-pip.py

and then install the packages by typing

pip install --user scipy
pip install --user scikit-image

Basic Usage

A fitting font-end are provided in the form of a python script

scripts/fit.py

which provides extensive help with the --help command line option.

The results of likelihood scans obtained with this script can be plotted using

scripts/plotscan.py

The results of individual fits (parameter pulls) can be with this script can be plotted using

scripts/plotpulls.py

There resulting files are .tex files that encode the graphics using TikZ and can be rendered with pdflatex or xelatex.

You can investigate log files with scripts/logdoc.py.

Advanced Usage

This section includes some examples on how this package can be used to perform some common tasks.

Liklihood scan

Create job definitions with

python scripts/fit.py --input workspace.root --fit --data asimovData --scan mu 100 0 5 --writeSubmit scan

Use some batch submission of your choice to submit the jobs defined in

scan/jobs.txt

After your jobs finish, plot a simple, 1D likelihood scan with

scripts/plotscan.py -i red scans/*.txt -o my1dscan.tex --atlas Preliminary --poi mu

Ranking

Create breakdown job definitions with

python scripts/fit.py --input workspace.root --fit --data asimovData --breakdown "ATLAS_*" --breakdown "theo_*" --breakdown "gamma_*" --writeSubmit breakdown

This will group parameters together, one group per --breakdown argument, and evaluate their combined influence of the POI(s) by calculating the difference in quadrature between a fit with these parameters set constant and the full fit.

If you are instead interested in evaluating a ranking by fixing the nuisance parameters to their +/-1 sigma values and re-fitting to evaluate the influence on the central value of the POI, you can use the impact mechanism. Similarly, you can create impact job definitions with

python scripts/fit.py --input workspace.root --fit --data asimovData --impacts "ATLAS_*" "theo_*"  "gamma_*" --writeSubmit impacts

If you want to derive postfit impacts, you need to run in two steps like this:

python scripts/fit.py -i workspace.root --data asimovData --poi mu --fit --write-workspace postfit.root
python scripts/fit.py -i postfit.root --poi mu --data asimovData --impacts 'norm_bkg' 'gamma_stat_*' --writeSubmit impacts 

Use some batch submission of your choice to submit the jobs defined in

breakdown/jobs.txt
impacts/job.txt

If you don't have a batch system at hand, you can also just execute these .txt files as bash scripts to run the jobs sequentially on your machine.

After your jobs finish, plot a full breakdown and impact plot including parameter pulls with

scripts/plotpulls.py -i red "breakdown/*nominal*.json" --impacts mu "breakdown/*.json" --range -2 2 --numbers --atlas True --output breakdown.tex

or

scripts/plotpulls.py -i red "impacts/*nominal*.json" --impacts mu "impacts/*.json" --range -2 2 --scaleimpacts 5 --numbers --atlas True --output impacts.tex

Once done, you can compile these files with pdflatex to look at your plots.

Pruning

As an example, you can try the pruning code on a Higgs 5XS workspace. The directory /afs/cern.ch/user/r/rabalasu/public/test_prune contains the 5XS 80ifb workspace (pre-fit and post-fit) and the corresponding hesse and fitresults as root files. To test the pruning code, copy its contents to test. The first step would be to obtain the following inputs for the pruning code, the postfit workspace, and the fitresult. If the workspace contains more POIs than those that should be profiled, make sure to specify only those that should be profiled. You can skip this step for the example as they are already available.

python scripts/fit.py --poi r_ggF r_VBF r_WH r_ZH r_ttH --input test/WS-Comb-5XS_80ifb.root --output test/WS-Comb-5XS.txt --workspace combWS --data combData --no-findSigma --hesse --writeResult --make-snapshots --write-workspace prune_test/WS-Comb-5XS_80ifb_postFit.root

To perform the prune run the order_NPs.py script. Running the order_NPs script with --writeSubmit options creates a txt file whose lines split the rank-finding into multiple jobs.

python scripts/order_NPs.py --pois r_ggF,r_VBF,r_WH,r_ZH,r_ttH --fitResult test/WS-Comb-5XS_80ifb_fitresult.root --writeSubmit test/joblines_5XS.txt --jobTime 10 --output test/order_NPs.txt

This creates the joblines file test/joblines_5XS.txt. You can run these lines in parallel which will write out the nuisance parameters and their ranks. Once all of them have finished, you will see the ranks for the nuisance parameters witten in test/order_NPs_*.txt. You can then run the pruning procedure with the following line,

python scripts/prune_NPs.py --input test/WS-Comb-5XS_80ifb_postFit.root --fitResult test/WS-Comb-5XS_80ifb_fitresult.root --pois r_ggF,r_VBF,r_WH,r_ZH,r_ttH --snapshot-name prune_combData_1pct --order test/order_*.txt --output test/WS-Comb-5XS_80ifb_postFit_prune.root --write-list test/pruned_NPs.txt

This creates a workspace WS-Comb-5XS_80ifb_postFit_prune.root that contain the snapshots prune_combData_1pct that contains the nuisance parameters identified for their respective thresholds set to 1% of max. change in the hesse error of any POI. The snapshot can be loaded before subsequent fits to fix the pruned nuisance parameters.