RDataFrame explicitly see how often event loop was executed

[gollumben]

Dear ROOT developers,

is there an explicit way with which I can see how often an event loop was executed? This would be super useful for me to see for debugging purposes. Maybe there is a way to output a string every time the event loop is executed?

Using RDataFrame with ROOT 6.14 and pyROOT.

Cheers,
Ben

[Danilo]

Hi Ben,

perhaps the simplest would be too do something like this:

ROOT::RDataFrame df(myTreename, myFileName);
auto d = df.Filter([](ULong64_t e){if (0ULL == e) std::cout << "Running evtloop" << std::endl;},{"rdfentry_"});
// now start to work on 'd'. At every loop, at he first event, you'll have a message.

We plan to add a “debug mode” to rdf and we’ll keep this topic in mind during the design: thanks for the post!

Cheers,
D

[gollumben]

Hi D,
thank you for this great idea! I tried to make a little minimum running example and unfortunately it is not working. The error message is

root [0] 
Processing test.C...
terminate called after throwing an instance of 'std::runtime_error'
  what():  Unknown column: rdfentry_

Could you kindly help me?

Cheers,
Ben
test.C (1.6 KB)

[eguiraud]

It might still be tdfentry_ (with a t) in v6.14 (v6.16 accepts both tdfentry_ and rdfentry_).

[gollumben]

Hi thank you, that solved the issue in C++. However, I cannot get it to run in Python. I defined a function there:

showWhenRunningEventLoop = '''
  bool showWhenRunningEventLoop(ULong64_t e){
    if (0ULL == e) std::cout << "!!!!! Running evtloop" << std::endl;
    return true;
  }
'''
ROOT.gInterpreter.Declare(showWhenRunningEventLoop)

I am trying to call it via

dataFrame = dataFrame.Filter("showWhenRunningEventLoop(tdfentry_)")

Unfortunately, when trying to define a variable next, which works without the above filter, ROOT replies to me that

ROOT::RDF::RInterface<ROOT::Detail::RDF::RJittedFilter,void> ROOT::RDF::RInterface<ROOT::Detail::RDF::RJittedFilter,void>::Define(experimental::basic_string_view<char,char_traits<char> > name, experimental::basic_string_view<char,char_traits<char> > expression) =>
The main RDataFrame is not reachable: did it go out of scope? (C++ exception of type runtime_error)

Do you know a way to fix this?

I also tried calling

dataFrame = dataFrame.Filter("showWhenRunningEventLoop",("tdfentry_"))

but this did not work either…

Cheers,
Ben

[Danilo]

Hi,

in Python that would look like this:

import ROOT 

d = ROOT.ROOT.RDataFrame(2)
c = d.Filter('if(rdfentry_ == 0) {cout << "Running evtloop" << endl; return true; } return false; ').Count()
print (c.GetValue())

(in root 6.14 rdfentry_ would be tdfentry_)

Cheers,
D

[gollumben]

Hi thanks,

I am using following code now:

ROOT.gInterpreter.Declare("int counterEventLoop = 0;")
eventLoopCounter = dataFrame.Filter('if(rdfentry_ == 0) {cout << "Running evtloop " << ++counterEventLoop << endl; return true; } return false; ').Count()

This now additionally gives me information on how often the event loop was ran. Thank you for all of your help!

[Danilo]

excellent!
Just remember not to go multithreaded with ROOT.ROOT.EnableImplicitMT() before making that counter atomic

Cheers,
D

[gollumben]

Thank you, that is helpful advice!

Now, I did a minimum running example, to test whether this is working well. Please find it attached. Essentially, I think it should be executing the event loop three times, if I understand RDataFrame correctly. So, I expect to see

Running evtloop 1
15.0
Running evtloop 2
15.0
Running evtloop 3
15.0

However, the output for me is

Running evtloop 1
15.0
15.0
15.0

I am sorry for bothering you again. What am I doing wrong?

minimumRunningExample.py (2.0 KB)

[Danilo]

Hi,

the Filter which prints to screen the event loop index is attached to the eventLoopCounter node.
You should attach the subsequent Defines to it rather than to dataframe.

Cheers,
D

[gollumben]

Thank you for clarifying this! Silly me!

Cheers,
Ben