Analysis in SPRACE

Available Datasets:

All datasets available in SPRACE (which are AOD or AODSIM) can be found in this link. For 2012, we have available the following real data datasets:

Single Muon
Double Electron
MET

Run2012 datasets can be found in this link

Summer12 MC datasets can be found in this link

JSON files

JSON files for 2012 Runs at 8 TeV can be found in this link.

Analysis steps

These links should be useful.

Online list of runs, triggers, etc.

General Analysis Strategy

In general, we advocate the following strategy:

Download datasets to SPRACE (optional)
Skim on basic reconstructed quantities // trigger bits. Run on GRID with CRAB. Save at SPRACE
Make Pattuples contataining everything you need for your analysis. Run on these using Condor. Save at SPRACE
Make basic ROOT ntuples containing very basic information for optimization // plots. Run on these at the interactive access server and/or your laptop.

In this way, we break the analysis in a hierarchical way: run on large datasets in the GRID, make smaller datasets to run on SPRACE, make Pattuples / ntuples to run in your computer. I think this is the more efficient strategy.

Skimming

Get the most recent JSON file for the link above
If you have already run on some data, do the difference in between the data you've already run upon and the new data with:
```
   compareJSON.py --sub <mostRecent.json> <dataAlreadyUsed.json> <fileForNewDataOnly.json>
   
```
Setup a CRAB job with the file for the new data only. In this example, we're running on the /MET/Run2012A-PromptReco-v1/AOD with the rsanalyzer_JetMET_skimming_Run2012A_cfg.py configuration file. We're setting up a task with around 75 jobs, and we will copy the output to the remote directory /MET_Run2012A-PromptReco_v1_2012May10, which lives in srm://osg-se.sprace.org.br:8443/srm/managerv2?SFN=/pnfs/sprace.org.br/data/cms/store/user/yourUserName/MET_Run2012A-PromptReco_v1_2012May10. Naturally, you have to setup these values for the ones you want.
```
[CRAB]
jobtype = cmssw
scheduler = glite
use_server = 0

[CMSSW]
datasetpath=/MET/Run2012A-PromptReco-v1/AOD
pset=rsanalyzer_JetMET_skimming_Run2012A_cfg.py
total_number_of_lumis=-1
number_of_jobs = 75
lumi_mask=fileForNewDataOnly.json
get_edm_output = 1

[USER]
copy_data = 1
return_data = 0
storage_element = T2_BR_SPRACE
user_remote_dir = /MET_Run2012A-PromptReco_v1_2012May10
ui_working_dir = myWorkingDirName

[GRID]
ce_white_list = T2_BR_SPRACE
   
```
Do the usual CRAB thing to get the output, but you also want the final report. This will produce a JSON file which resides in myWorkingDirName/res/lumiSummary.json. This file represents exactly the data over which you ran over, taking into account failed jobs, blocks of data which were not yet available, etc.
```
crab -status -c myWorkingDirName
crab -getoutput -c myWorkingDirName
crab -report -c myWorkingDirName
   
```
To get the amount of luminosity that you ran over, use the lumiCalc2.py script:
```
lumiCalc2.py -b stable -i lumiSummary.json overview
   
```
You should add this data to the set of data you already ran over. You do this by merging the JSON files. The syntax is:
```
mergeJSON.py previousData.json dataYouJustRanOver.json --output=totalData.json
   
```

The following picture shows this process schematically:

What should I use for this skimming step?

Naturally, it depends on your specific analysis channel. Remember that the goal is to separate the analysis hierarchically - run over large datasets using the GRID, preselect/reduce them to more manageable sizes, bring them to SPRACE and run the rest of the analysis more or less locally. If you make a very complicated preselection in the GRID, it starts to become comparable to make the whole analysis there, and defeats the whole idea. So, some general points:

You can skim on trigger bits. To do that, you use the TriggerResultsFilter module. You can see an example of trigger-based skimming in this link.
You can skim on basic RECO quantities. You use normal EDFilters for that. For inspiration, you can look at the official CMS skimming page. Notice that those are the official skims - what we are doing here is mimicking that scheme.

PATtuple making

In CMS we use the Physics Analysis Toolkit (PAT) to steer our analyses. It is a set of standard EDModules and configuration files that act as building blocks for you to build you analysis.