RDataFrame `GetBranchNames: unsupported branch type`

[schneiml]

Hi all,

I just tried to take a loook at analyzing some data using RDataFrame again, and hit the following problem:

root [1] auto df = ROOT::RDataFrame("MEs", "dqmio.root");
root [2] *df.Count()
(unsigned long long) 840840
root [3] auto only_CSC = [](TString name){ return name == "CSC/CSCOfflineMonitor/Occupancy/hOStripsAndWiresAndCLCT";};
root [4] auto csc_df = df.Filter(only_CSC, {"name"});
Error in <TRint::HandleTermInput()>: std::runtime_error caught: GetBranchNames: unsupported branch type

This is using ROOT 6.14/08, provided as default on lxplus7. It used to work in ROOT 6.12/07 (provided by CMSSW), using the old names:

root [0] auto df =  ROOT::Experimental::TDataFrame("MEs", "dqmio.root");
root [1] *df.Count()
(unsigned long long) 840840
root [2] auto only_CSC = [](TString name){ return name == "CSC/CSCOfflineMonitor/Occupancy/hOStripsAndWiresAndCLCT";};
root [3] auto csc_df = df.Filter(only_CSC, {"name"});
root [4] *csc_df.Count()
(unsigned long long) 13347

One of the branches contains TH2F objects, is this the problem? The error message is not very helpful…

You can check out the file here:
https://cernbox.cern.ch/index.php/s/CcgXcCJMweaaTck

It would be fun to look at this data using RDataFrame, given it will need some non-trivial filtering and operations applied to make sense.

Cheers,

Marcel

---

ROOT Version: 6.14/08
Platform: CC7 (lxplus7)
Compiler: linuxx8664gcc

---

[eguiraud]

Hi Marcel,
well done switching ROOT versions, v6.12 is a lot of time in RDF years
If you can, you should use v6.16 (available on lxplus7 as /cvmfs/sft.cern.ch/lcg/views/LCG_95/x86_64-centos7-gcc8-dbg/setup.sh) as there is also an important gap in performance, features and bugfixes in RDF between v6.14 and v6.16.

Now, about your specific issue: indeed it seems RDF (both v6.14 and v6.16) is not able to operate on your value branch, which contains a TH2. The technical reason is that we added some extra features since v6.12 which do not support TBranchObjects.
I opened a jira ticket, you can keep track of the progress on the issue there (and you should ping us there if you see no progress in the next 3-4 weeks, and this is a blocker for you).

Cheers,
Enrico

[eguiraud]

Hi,
it seems that the problems with TBranchObjects go deeper than RDF. Here’s a PR that seems to fix the RDF-related issues, but I can’t give a time estimate for a fix to the deeper problems.

Is it an option to move away from TBranchObjects and use e.g. TBranchElements instead?

Cheers,
Enrico

[schneiml]

Ciao Enrico,

thanks for the quick response.

For the urgency of this in general: We don’t really use RDataFrame currently, this is just for my personal experiments and I thought about using it in future versions of our tools. So not urgent, but if we can’t handle TH1/TH2 objects and do non-trivial operations on them (fitting?), we can’t use it for what I planned.

For TBranchElements: I don’t know much about the details here. I wrote the file using the most naive code I could come up with. If you can suggest a better way to write a TTree of histograms, I’d be interested.

The code I used to write the tree is [1], I do the round-trip via JSON and SQLite mostly because I had that format anyways and could not find an easy way to write a TTree of objects in Python…

Thanks,

Marcel

[1] https://github.com/cms-sw/cmssw/compare/master...schneiml:dqm-root-to-sqlite#diff-6b88da977b5e099ad42379a350fba3ceR249

[eguiraud]

Hi,
for some reason passing splitlevel 0 produces a TBranchObject instead of a TBranchElement – I’m not sure if it’s a bug or a feature, I’ll investigate:

TH2F *value;
TTree t;
t.Branch("value", &value,128*1024, 0)->IsA()->GetName(); // returns TBranchObject
t.Branch("value", &value,128*1024)->IsA()->GetName(); // returns TBranchElement

Could you try removing the splitlevel=0 and see if that makes TTreeReader (and therefore RDataFrame) happy?

P.S.
If it can be useful in the future, RDataFrame supports reading SQLite files directly with MakeSQLiteDataFrame.

[schneiml]

Hi Enrico,

I just re-generated the root file without that last 0 (I think I stole that from the CMSSW code that originally wrote the data), and things look a lot better. Now using 6.16/00, I get the same output as with 6.12 before.

Now I need to actually make some plots to see if things work correctly…

For reading the SQLite data directly: That was the original plan, but for the beginning, the ROOT file is a lot smaller and is also readable with older ROOT versions, while FromJSON is another rather recent feature. I might give that a try at some point.

Thanks a lot,

Marcel

[eguiraud]

Cool, let us know how that goes so we can mark this thread as solved

[eguiraud]

In case someone ends up here: reading TBranchObjects with RDF and TTreeReader was fixed in v6.22/02 or v6.24.