BruFit

A RooFit based event based maximum likelihood fitting package

The purpose of this package is to add to the RooFit package to allow analysis of hadronic physics scattering reactions.

The main feature is a PDF class (RooHSEventsPDF) that allows calculation of normalisation integrals from Monte Carlo detector simulations which allow the acceptance of detector systems to be corrected for when extracting obervables.

In addition the (RooComponentsPDF) class provides caching of these integrals for fast evaluation when the PDF is a sum of products

     A1(x_1)*B1(p_1) + A2(x_2)*B2(p_2) + ....

where x_i are measured variables and p_i are the parameters of interest. For example

C*cos(2*phi) +  D*sin(2*phi)

Has

A1 = cos(2*phi); x_1 = phi; p_1 = C; A2=sin(2*phi); x_2=phi; p_2=D 

The idea is that more complex fits should not require more complex code and there are components for splitting data (e.g into energy bins); running similar fits in parallel via PROOF or a batch farm, which require minimal extra code.

Weights can be used in the fits and can be created using the RooStats sPlot class.

A Markov Chain Monte Carlo implementation based on Metropolis Hastings is implemented and can provide robust (although not optimal) minimisation on fits theat minuit may struggle to find a global minimum.

Installation

Prerequisites

ROOT with RooFit, Proof, Mathmore (if using Legendre polynomials). Currently tested on 6.20, ...6.14 know to fail.

get and compile code

git clone https://github.com/dglazier/brufit.git

cd brufit

setenv BRUFIT /path/to/here (or setenv BRUFIT $PWD)

mkdir build

cd build

cmake ../

make install

Note to install the pcm file you may have to run the last two steps again. You can check if $BRUFIT/lib/libbrufit_rdict.pcm exists.

cmake ../

make install

alias brufit root $BRUFIT/macros/LoadBru.C

Data

Data should be in the form of a ROOT TTree with branches that are usually double but can be int for categories, e.g. a polarisation state. If you are using weights and need an event ID branch this should also be made double so it can be read into RooFit dataset.

Basic usage

 > brufit
 root [1] FitManager fm
 root [2] fm.SetUp().SetOutDir("out/"); //Put results files in out/
 root [3] fm.SetUp().LoadVariable("phi[-3.1416,3.1416]"); //phi is a variable in the data tree
 root [4] fm.SetUp().FactoryPDF("EXPR::amplitude('1+A[0,-1,1]*cos(2*phi)',phi,A)"); //Fit a cos2phi distribution
 root [5] fm.SetUp().LoadSpeciesPDF("amplitude");//add to the total fit PDF 
 root [6] fm.LoadData("treeName","fileName.root"); //set data (ROOT tree)
 root [7] Here::Go(&fm); //run the fit

Tutorials

sPlotSimple

get the files

  cp -r $BRUFIT/tutorials/sPlotSimple .
  cd sPlotSimple

If running with Jupyter (Recomended)

start a notebook. Note the tutorials are written in python3 kernels.

  root --notebook  or jupyter-notebook

And open sPlotSimple.ipynb

You can also try the sPlotSimpleBins for an example of splitting the data into energy bins before making several fits.

If Running in ROOT interactive

First make some data

  root 'Model1.C( "Data.root" )'

Run

brufit FitHSSimple.C

and

brufit FitHSSimpleBins.C

Performing fits to sinusiodal distributions with weights

  cp -r $BRUFIT/tutorials/WeightedObservable .
  cd WeightedObservable

If running with Jupyter (Recomended)

start a notebook. Note the tutorials are written in python3 kernels.

  root --notebook  or jupyter-notebook

And open sPlot.ipynb

Once you have found weights you can perform the weighted fit from FitWithEventsPDF.ipynb

You can also try using simulated data to give your sPlot Signal shape in sPlotWithSimulatedPDF.ipynb. And then try these weights in FitWithEventsPDF

Finally you can try splitting the Fit into Eg bins, running seperately on PROOF then plotting the result parameters as a function of Eg with FitWithComponentsPDFAndSplitBins.ipynb.

A faster more optimised method using RooComponentsPDF is given in FitWithComponentsPDF.ipynb

If Running in ROOT interactive

See the README in tutorials/WeightedObservable