130th ROOT Parallelism, Performance and Programming Model Meeting
We will meet in the ROOT team Zoom room:
https://cern.zoom.us/j/97374667082?pwd=YjJDcFlOMXVTam9hRVBQb0lqRFgrUT09
- no simple solution for auto-binning in distRDF, but should take a better look at what PROOF did
- possible solutions require synchronization across processes (the more expensive the more histograms there are) or a pilot run that stops after each histogram has seen 100 entries to estimate their bin limits
- no sufficient need, at least at the moment, to justify the complexity or performance loss of such solutions, especially given that auto-binning is only useful for exploratory analyses, and exploratory analyses can usually be run at smaller scales (hence locally, where auto-binning is available)
- auto-naming of varied histograms: leave title alone (users might have customized it with latex or with axes names), have different names to simplify saving to ROOT files
- remove the `:` in variation names so histogram names that contain the variation name are still valid identifiers (e.g. for PyROOT's `file.histo_ptUp`)
- given good default names, unclear whether we can make customization more ergonomic than simply looping over the varied histograms and calling `h.SetName(h.GetName().replace(syst, ...))`
- sample labels might have to be dictionaries of metadata rather than a single string (but if they are std::maps all values have to be the same type, so probably strings, which is limiting...)
- unclear whether there is a usecase for labels attached to friends, most if not all metadata you would want to attach is event-wide, so it spans the friends as well, so it can be attached to just the main chain