Loop over interation to create gamma-gamma matrix using RDataFrame

joa · January 18, 2022, 1:58pm

Dear all,

some info…

ROOT Version: 6.23/01
Platform: ubuntu 20.04
Compiler: g++ (Ubuntu 9.4.0-1ubuntu1~20.04) 9.4.0

then the problem

I’m trying to understand how to write something like
(gammas is a vector stored in a tree)

for(auto &it : gammas){
 for(auto &it2 : gammas){
 if(it2!=it){
 A2DMatrix.Fill(it,it2);
}
}

using a RDataFrame but can not quite figure out how to loop over interations in an event using RDataFrames.

best regards

Joa

couet · January 18, 2022, 2:16pm

I think @eguiraud can help you.

eguiraud · January 18, 2022, 2:32pm

Hi @joa ,
in first approximation you can write exactly that code:

TMatrix A2DMatrix;

auto fillA2DMatrix = [&] (const ROOT::RVec<float> &gammas) {
  for(auto &it : gammas){
    for(auto &it2 : gammas){
      if(it2!=it){
        A2DMatrix.Fill(it,it2);
      }
    }
 }
};

df.Foreach(fillA2DMatrix, {"gammas"});

This only works on a single thread, of course.
Depending on your exact usecase we can probably also implement something that works in a multi-thread run, but I hope this can get you started.

Cheers,
Enrico

joa · January 18, 2022, 2:46pm

Hi,

Ok. So for large data sets I’ll be better off with a TSelector and proof…?
This is not threadsafe because of the Fill in TMatrix, correct?

cheers

Joa

eguiraud · January 18, 2022, 3:01pm

Correct.

There are some simple ways to make this thread-safe, for example (again, just to give you an idea, the best solution depends on your exact problem):

ROOT::EnableImplicitMT();
ROOT::RDataFrame df(...);
std::vector<Matrix2D> matrices(df.GetNSlots()); // a vector of nThreads matrices
df.ForeachSlot([&matrices](unsigned int slot, const ROOT::RVec<float> &gammas) {
  for(auto &it : gammas){
    for(auto &it2 : gammas){
      if(it2!=it)
        matrices[slot].Fill(it,it2);
    }
 }
}, {"gammas"});
// ...merge matrices...