tth-analyze-mem

Matrix element method for ttH, H->tautau channel

Building instructions

Clone the library into $CMSSW_BASE/src/tthAnalysis/tthMEM:

git clone https://github.com/HEP-KBFI/tth-analyze-mem.git $CMSSW_BASE/src/tthAnalysis/tthMEM

In order to build the whole project, you also need to install numerical integration library VAMP by following these steps:

mkdir $CMSSW_BASE/VAMP
wget http://whizard.hepforge.org/oldsrc/vamp-2.3.0.tar.gz -P $CMSSW_BASE/VAMP
tar zxvf $CMSSW_BASE/VAMP/vamp-2.3.0.tar.gz -C $CMSSW_BASE/VAMP
mkdir -p $CMSSW_BASE/VAMP/vamp-2.3.0/prefix/share/doc/vamp
cd $CMSSW_BASE/VAMP/vamp-2.3.0
./configure --prefix=$CMSSW_BASE/VAMP/vamp-2.3.0/prefix
make -j4
make install
cp $CMSSW_BASE/VAMP/vamp-2.3.0/prefix/lib/* $CMSSW_BASE/lib/$SCRAM_ARCH
cp $CMSSW_BASE/src/tthAnalysis/tthMEM/test/vamp.xml $CMSSW_BASE/config/toolbox/$SCRAM_ARCH/tools/selected/vamp.xml

Next up try to set up with scram setup vamp and verify the installation with scram tool info vamp. You should see output along the lines of

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Name : vamp
Version : 2.3.0
++++++++++++++++++++
SCRAM_PROJECT=no
LIB=vamp
LIBDIR=.../VAMP/vamp-2.3.0/prefix/lib
INCLUDE=.../VAMP/vamp-2.3.0/prefix/include/vamp
USE=f77compiler

You can check other versions of VAMP here.

Finally, build it with scram b -j8 as usual.

Running an example

At the time of writing, the project includes one executable, runMEM, which takes a Python configuration file as its argument. The program needs .root file(s) containing selected events; an example configuration file is provided in data/runMEM_3l1tau_cfg.py and example data file in data subdirectory. The output file is the original file + signal and background probabilities found with the MEM.

In order to run the MEM via SLURM, one has to call python createJobs_3l1tau.py in any directory and follow instructions on screen. Note that this assumes you have run the tth analysis: https://github.com/HEP-KBFI/tth-htt/ (with select_root_output set to True).

The project also includes some unit tests, which can be run with scram b -j8 runtests after building the project.

Disabling logging

If you want to gain in speed, you are advised to build the project w/o no logging whatsoever. This vcan be achieved w/ the following command

USER_CPPFLAGS="-DDISABLE_LOGGING" scram b -j8

Valgrind commands

Here we provide some example valgrind commands which help to debug the application. Of course, in order to benefit from the Valgrind bundle, one should definitely build the project w/ debug symbols enabled:

scram b -j8 USER_CXXFLAGS="-g -Wuninitialized"

Commands

Memory leak detection:

valgrind --tool=memcheck `cmsvgsupp`                               \
--suppressions=$CMSSW_BASE/src/tthAnalysis/tthMEM/data/tthMEM.supp \
--leak-check=yes                                                   \
--show-reachable=yes                                               \
--num-callers=20                                                   \
--track-fds=yes                                                    \
--track-origins=yes                                                \
--log-file="valgrind.log"                                          \
runMEM_3l1tau python/runMEM_3l1tau_2016_cfg.py                     \
&> out.log

The second line suppresses leaks from 3rd party libraries linked to runMEM_3l1tau. If you only see Python-related leaks, then you're good.

Memory consumption:

valgrind --tool=massif                         \
--depth=40                                     \
--time-stamp=yes                               \
--time-unit=ms                                 \
--threshold=0.1                                \
runMEM_3l1tau python/runMEM_3l1tau_2016_cfg.py \
&> out.log

Callgraph:

valgrind --tool=callgrind                      \
runMEM_3l1tau python/runMEM_3l1tau_2016_cfg.py \
&> out.log