Skip to content

Conversation

vincecr0ft
Copy link
Contributor

The changes from the wreckage of the last few months commits have been salvaged and pieced bit by bit back into the main RooFit rootbench.

A second code splits the RooFit tests into Binned and UnBinned cases. (this reflects the differences LHCb/Babar style problems based on custom PDFs built in standalone RooFit (analytical form is known) to the ATLAS/CMS style (where the form of the PDF is estimated using a morphing scheme within HistFactory). The UnBinned code was written independently of recent updates to the code and therefore may need some optimisation (removing the profile timers etc).

The binned test scales in both #channels and #bins each with 1-3 CPUs assigned to handle the components of the evaluation. Hesse requires that Migrad is run first and Minos that Hesse runs. A snapshot taken before the first minimisation is applied to reset the parameter values before the timed minimisation. As such minos, by definition, will take longer to run then Hesse and Hesse longer than Migrad as the execution of each is included however only the final minimisation is timed.

@vincecr0ft
Copy link
Contributor Author

vincecr0ft commented Dec 19, 2017

locally I run into this error:

libc++abi.dylib: terminating with uncaught exception of type RooFit::BidirMMapPipe_impl::BidirMMapPipeException: pipe: Too many open files
Abort trap: 6

Towards the end of the operations. This has already been mitigated by restricting profiles to a fixed number of iterations. However when including both #Channel and #Bin scaling tests the problem returns. (right at the end)

@vincecr0ft
Copy link
Contributor Author

With respect to the comment above. I don't remember this being an issue when each workspace was called individually (instead of now where everything is just called "workspace.root"). I also know that RooFit has trouble letting go of objects (in python sometimes we have to explicitly 'disown' workspace objects created by RooFit in order to export code)

@coveralls
Copy link

Coverage Status

Coverage remained the same at 0.0% when pulling bbe98be on roofit-dev:LongerBinnedTests into 5366fe5 on root-project:master.

@oshadura
Copy link
Contributor

@vincecr0ft, thank you so much! Great job! tests are a bit long (RT), so I would like to add them to separate "long RT benchmark group" maybe..

@oshadura
Copy link
Contributor

@vincecr0ft can I ask you rebase please? :)

@coveralls
Copy link

Coverage Status

Coverage remained the same at 0.0% when pulling 8d4c9aa on roofit-dev:LongerBinnedTests into b03d819 on root-project:master.

@oshadura
Copy link
Contributor

oshadura commented Feb 1, 2018

Closing, was merged already in a different PR

@oshadura oshadura closed this Feb 1, 2018
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants