Introduction

First of all you have to install Professor, eigen3, matplotlib and yoda on your notebook: https://professor.hepforge.org/docs.sphinx220/index.html

General information about how to start tuning can be found here: https://twiki.cern.ch/twiki/bin/view/AtlasProtected/MCTuningGroup#Getting_started_with_tuning

For our excercise we start from already produced test samples used for the A15 tune.

First, get the output of the mc runs from lxplus: /afs/cern.ch/work/a/aknue/public/TestTune_A15_Tunathon/

cd TestTune_A15_Tunathon

This folder contains three items:

• mc -> folder that contains all parameter variations
• refdata -> folder that contains all reference data from Rivet
• WeightsTunathon -> txt file that lists all plots included in the tune with a corresponding weight (default: all set to 1.0)

The folder "mc" itself contains 500 numbered folders, that have the following contents:

• A file "used_params" which tells you what parameter set was used for the particular run.
• A yoda file in which you can find the output of the Rivet routines.

To save time we basically start with step 5 and 6 of the instructions given in the twiki above and skip the sampling of the parameters and the MC generation.

An overview over all available rivet routines can be found here: https://rivet.hepforge.org/analyses

The analyses used in the A15 tunes are listed below.

Make the run combinations

As you can see in the "mc" folder, we have 500 set of parameters, called "runs" in the following.

First you have to make so-called run combinations. The idea is that you use either the full set of available parameters, or a subset of them.

In order to use all available parameters, you do the following (assuming that Professor and Yoda are already set up properly):

prof2-runcombs mc/ --pname=used_params 0:1 -o runcomb.dat

Then you should get an output file called runcomb.dat which contains all foldernames that should be used in the tuning step. The "0" here means, that no runs are disregarded.

If you just want to use a subset of runs, just do the following:

prof2-runcombs mc/ --pname=used_params 50:200 -o runcomb_subset.dat

The "50" means now that 450 out of the 500 runs should be used, and the "200" is the number of combinations which should be produced from these runs. If you check runcomb_subset.dat you will see that you have now 200 lists of folders stored in the file.

Interpolate the generator response with Professor

For the interpolation step you have different options. You can find them all with:

prof2-ipol --help

So as you can see, there are different options for the interpolation function etc.

For the default settings you just need to run:

prof2-ipol mc/ --pname=used_params

This step will take a couple of minutes (10-15, so time enough to get coffee).

As output of this step, you will get the file ipol.dat.

Perform the tuning

Now for the actual tuning step, do the following:

prof2-tune -d refdata/ --wfile=WeightsTunathon ipol.dat -r mc/ -o tunesTest --debug --filter

The best set of parameters you can find now in "tunesTest/results.txt".

Envelopes and sensitivity plots

In order to make the envelopes do:

prof2-envelopes mc/ refdata/ --pname=used_params

You get the output stored in the new "envelopes" folder.

Make sensitivity plots:

prof2-sens ipol.dat --grad

You get the output stored in the new "sensitivities" folder. If you use --cmap instead of --grad (one-dimensional), you will get colour map plots. This step has however a bit longer running time.

Now you can redo the different steps by for example increasing/decreasing the weights for the different distributions and see if the result improves. You can also play with different interpolation functions, set limits on the parameter ranges etc.

Tuning with/without ttbar data

Now we can also test the difference in the tuning results with/without the ttbar data included. Just set the weights for the ttbar histograms to zero for testing.

Eigenvariations

I did not get to this part unfortunately

-- AndreaKnue - 2017-12-12