rx-data 0.2.2.dev630

Project with lists of LFNs and utilities needed to download filteres ntuples

[TOC]

$R_X$ data

This repository contains:

• Versioned lists of LFNs
• Utilities to download them

for all the $R_X$ like analyses. For instructions on how to:

• Produce new ntuples with friend trees
• Downloading filtered ntuples from the grid
• Merging data ntuples
• Copying ntuples from cluster to laptop
• Outdated instructions that hasn't been removed yet

Check this.

Below are the instructions on how to access data from EOS.

Installation

To install this project run:

pip install git+ssh://[e-mail]:7999/rx_run3/rx_data.git

The code below assumes that all the data is in ANADIR. If you want to use the data in EOS do:

export ANADIR=/eos/lhcb/wg/RD/RX_run3

preferably in ~/.bashrc.

How the the code makes the ROOT dataframes

When creating datframes, the code will:

• Check the directories where the ROOT files are
• Make lists of paths
• Create dictionaries with these paths, split into samples and save them in yaml files. Each yaml file is associated to a different friend tree or the main tree.
• For a given sample, pick up the lists of paths from the yaml files and create a JSON file
• Use the JSON file to make the ROOT dataframe by using from_spec RDataFrame's method

Accessing ntuples

Once

from rx_data.rdf_getter     import RDFGetter

# This picks one sample for a given trigger
# The sample accepts wildcards, e.g. `DATA_24_MagUp_24c*` for all the periods
gtr = RDFGetter(
    sample   = 'DATA_24_Mag*_24c*',
    tree     = 'DecayTree'              # This is the default, could be MCDecayTre
    trigger  = 'Hlt2RD_BuToKpMuMu_MVA')  # This should allow picking RK, Rkstar or noPID samples

# If False (default) will return a single dataframe for the sample
rdf = gtr.get_rdf(per_file=False)

# If True, will return a dictionary with an entry per file. They key is the full path of the ROOT file
d_rdf = gtr.get_rdf(per_file=True)

The supported triggers are:

Trigger	Usage
Hlt2RD_BuToKpMuMu_MVA	$R_K$ muon samples
Hlt2RD_B0ToKpPimMuMu_MVA	$R_{K^*}$ muon samples
Hlt2RD_BuToKpEE_MVA	$R_K$ electron samples
Hlt2RD_B0ToKpPimEE_MVA	$R_{K^*}$ electron samples
Hlt2RD_BuToKpMuMu_MVA_noPID	$R_K$ muon samples
Hlt2RD_B0ToKpPimMuMu_MVA_noPID	$R_{K^*}$ muon samples
Hlt2RD_BuToKpEE_MVA_noPID	$R_K$ electron samples
Hlt2RD_B0ToKpPimEE_MVA_noPID	$R_{K^*}$ electron samples

The way this class will find the paths to the ntuples is by using the DATADIR environment variable. This variable will point to a path $DATADIR/samples/ with the YAML files mentioned above.

In the case of the MVA friend trees the branches added would be mva.mva_cmb and mva.mva_prc.

Thus, one can easily extend the ntuples with extra branches without remaking them.

Sample emulation

Certain samples are not available, but they can be emulated from existing ones. E.g. $B_s \to J/\psi K^$ can be obtained through $B_d \to J/\psi K^$. This is configured in rx_data_data/emulated_trees/config.yaml as:

Bs_JpsiKst_mm_eq_DPC :          # This is the sample needed
  sample : Bd_JpsiKst_mm_eq_DPC # It will be replaced by this
    redefine :                  # where the changes will be in this section
    B_M    : B_M    + 87.23
    B_Mass : B_Mass + 87.23

Branches

The list of branches is:

Project	Channel	Link
$RK$	Electron	link
$RK$	Muon	link
$RK^*$	Electron	link
$RK^*$	Muon	link

Checking what samples exist as filtered ntuples in the grid

This is useful to avoid filtering the same samples multiple times, which would

• Slow down the analysis due to the large ammount of data needed to download
• Occupy more space in the user's grid

For this run:

from rx_data.filtered_stats import FilteredStats

fst = FilteredStats(analysis='rx', versions=[7, 10])
fst.exists(event_type='12153001', block='w31_34', polarity='magup')

This will check if a specific sample exist in the versions 7 or 10 of the filtering. Where these versions are the versions of the directories in rx_data_lfns/rx.

This will require access to the user's ganga sandbox through the GANGADIR variable. This should be improved eventually, ideally by integrating the filtering with the analysis productions pipeline.

Checking what samples exist as ntuples in ANADIR (locally)

For this run:

check_local_stats -p rk

	With PID	Without PID
$R_K$	here	here
$R_{K^*}$	here	here

Where the rows represent samples and the columns represent the friend trees. The numbers are the number of ntuples.

Multithreading

Multithreading with ROOT dataframes at the moment is dangerous and should be done only in a few places. To turn this on run:

nthreads = 3 # Or any reasonable number
with RDFGetter.multithreading(nthreads=nthreads):
    gtr = RDFGetter(sample=sample, trigger='Hlt2RD_BuToKpEE_MVA')
    rdf = gtr.get_rdf()

    process_rdf(rdf)

• Once outside the manager, multithreading will be off.
• One can use nthreads=1 to turn off mulithreading
• Negative or zero threads will raise exception.

Unique identifiers

In order to get a string that fully identifies the underlying sample, i.e. a hash, do:

gtr = RDFGetter(sample='DATA_24_Mag*_24c*', trigger='Hlt2RD_BuToKpMuMu_MVA')
uid = gtr.get_uid()

Identifiers for cluster jobs

When sending jobs to a computing cluster, each job will try to read the data. Thus, it will create the JSON and YAML files mentioned above. If two jobs run in the same machine, this could create clashes and failed jobs. To avoid this do:

from rx_data.rdf_getter    import RDFGetter

sample = 'Bu_JpsiK_ee_eq_DPC'
with RDFGetter.identifier(value='job_001'):
    gtr = RDFGetter(sample=sample, trigger='Hlt2RD_BuToKpEE_MVA')
    rdf = gtr.get_rdf(per_file=False)

i.e. wrap the code in the identifier manager, which will name the files based on the job.

Excluding datasets

One can also exclude a certain type of friend trees with:

from rx_data.rdf_getter     import RDFGetter

wih RDFGetter.exclude_friends(names=['mva']):
    gtr = RDFGetter(sample='DATA_24_Mag*_24c*', trigger='Hlt2RD_BuToKpMuMu_MVA')
    rdf = gtr.get_rdf(per_file=False)

that should leave the MVA branches out of the dataframe.

Defining custom columns

Given that this RDFGetter can be used across multiple modules, the safest way to add extra columns is by specifying their definitions once at the beggining of the process (i.e. the initializer function called within the main function). This is done with:

from rx_data.rdf_getter     import RDFGetter

RDFGetter.custom_columns(columns = d_def)

If custom columns are defined in more than one place in the code, the function will raise an exception, thus ensuring a unique definition for all dataframes.

Accessing metadata

Information on the ntuples can be accessed through the metadata instance of the TStringObj class, which is stored in the ROOT files. This information can be dumped in a YAML file for easy access with:

dump_metadata -f root://[e-mail]//eos/lhcb/grid/user/lhcb/user/a/acampove/2025_02/1044184/1044184991/data_24_magdown_turbo_24c2_Hlt2RD_BuToKpEE_MVA_4df98a7f32.root

which will produce metadata.yaml.

Run1/2 samples

For now these samples are only in the UCAS cluster and only the rare electron signal has been made available through:

from rx_data.rdf_getter12 import RDFGetter12

gtr = RDFGetter12(
    sample ='Bu_Kee_eq_btosllball05_DPC', # BuKee
    trigger='Hlt2RD_BuToKpEE_MVA',        # This will be the eTOS trigger
    dset   ='2018')                       # Can be any year in Run1/2 or all for the full sample

rdf = gtr.get_rdf()

this dataframe has had the full selection applied, except for the MVA, q2 and mass cuts.

Cuts can be added with:

from rx_data.rdf_getter12 import RDFGetter12

d_sel   = {
    'bdt' : 'mva_cmb > 0.5 & mva_prc > 0.5',
    'q2'  : 'q2_track > 14300000'}

with RDFGetter12.add_selection(d_sel = d_sel):
    gtr = RDFGetter12(
        sample =sample,
        trigger=trigger,
        dset   =dset)

    rdf = gtr.get_rdf()