2023:groups:smtools:start
Differences
This shows you the differences between two versions of the page.
| Both sides previous revision Previous revision Next revision | Previous revision | ||
|
2023:groups:smtools:start [2023/06/16 14:24] josh.mcfayden |
2023:groups:smtools:start [2023/06/19 11:11] (current) stefan.kiebacher |
||
|---|---|---|---|
| Line 20: | Line 20: | ||
| - Extending study from LH17 [[https://arxiv.org/abs/1803.07977|arXiv:1803.07977]] | - Extending study from LH17 [[https://arxiv.org/abs/1803.07977|arXiv:1803.07977]] | ||
| - Effective/comprehensive uncertainty band | - Effective/comprehensive uncertainty band | ||
| - | - Uncertainty recommendation using available tools [Josh] | + | - Uncertainty recommendation using available tools [Josh, Gianna] |
| - Setup 4-point variations of all combinations of shower and hadronisation model | - Setup 4-point variations of all combinations of shower and hadronisation model | ||
| - Tune these in a consistent way to get an uncertainty envelope | - Tune these in a consistent way to get an uncertainty envelope | ||
| Line 30: | Line 30: | ||
| - | ===Generator software and computing considerations [Friday 11h00-12h30 QCD room]=== | + | ===Generator software and computing considerations [Friday 11h00-12h30 Auditorium]=== |
| Slack channel: #gen-soft-comp | Slack channel: #gen-soft-comp | ||
| Line 41: | Line 41: | ||
| * Use of ML | * Use of ML | ||
| * Common frameworks | * Common frameworks | ||
| - | Interested people: Josh, Vinny, Andrea, Stefan, Saptaparna, Maria V., Sergio, Chris G, Christian P, Ana, Max, Peter M, Simon | + | Interested people: Josh, Vinny, Andrea, Stefan, Saptaparna, Maria V., Sergio, Chris G, Christian P, Ana, Max, Peter M, Simon, Sudeepan |
| **Agreed projects:** | **Agreed projects:** | ||
| - | - | + | - GPU/Vectorisation/HPC [Josh] |
| + | - Experience porting codes | ||
| + | - Workflows in experiments | ||
| + | - How would code be run in practise | ||
| + | - How to get/confirm allocations | ||
| + | - Sharing of resources between EXP and TH | ||
| + | - Computing performance [Josh] | ||
| + | - Benchmarking current code | ||
| + | - Also with examples from EXP | ||
| + | - Accounting in experiments | ||
| + | - Projections for future N(2,3)LO calculations | ||
| + | - Updates to interfaces [Josh] | ||
| + | - Interoperability of models | ||
| + | - Modular framework | ||
| + | - Multi-event API in MC generators | ||
| + | - Resampling | ||
| + | - Accessibility of tools [Josh] | ||
| + | - Containerisation and reproducibility | ||
| + | - Docker images, run cards, Rivet routines, Yoda files -> HepData | ||
| + | - Use these tools to make LH studies reproducible | ||
| + | - Can at Docker-based workflow on e.g. lxplus be documented to lower barrier to entry for MC studies? | ||
| + | - Make sure cards actually used by EXP are uploaded with TH prediction Yoda files. | ||
| + | - Make available for benchmarking (c.f. point 2.), including full chain in EXP | ||
| + | - MC Generation generic tool/interface | ||
| + | - Like MCPlots | ||
| + | - Event formats [Josh] | ||
| + | - New HDF5 standard <- LHEF/HepMC | ||
| + | - Easier event sharing, smaller disk footprint | ||
| + | - Need for more information in intermediate stages of generation? | ||
| + | - Sample sharing between experiments | ||
| + | - Joint/cross validation | ||
| + | - ML for MC [] | ||
| + | - Interoperability link to point 3. (swapping out parts of generation in modular framework) | ||
| + | - Discussion of exact vs approximate use of ML | ||
| + | - Resampling/reweighting [Josh] | ||
| + | - Reweighting and derivative-based optimisation | ||
| + | - Resampling to eliminate negative weights | ||
| + | - NN and cell-based tools | ||
| + | - Stress tests of these tools? | ||
| + | - Problem areas from TH PoV | ||
| + | - Validation at scale in EXP | ||
| ===ttW modelling studies [TBC]=== | ===ttW modelling studies [TBC]=== | ||
| Line 66: | Line 106: | ||
| Interested: Simon, Andrzej, Stefan K. | Interested: Simon, Andrzej, Stefan K. | ||
| + | |||
| + | |||
| + | ===Tuning Accord and Strategies=== | ||
| + | * What is currently done: | ||
| + | - Common framework for tuning exist, but no single script to run | ||
| + | - Nice to have a script, which automatically prints exactly what is done and with which flags | ||
| + | - Every generator team does tuning on their own | ||
| + | - Almost no documentation on the tunes provided | ||
| + | - Rarely publicly available weights | ||
| + | - Rarely publicly available tuning options (samples, interpolation order, flags used, ...) | ||
| + | - Often post-processing tune by eye | ||
| + | * Goals of the project: | ||
| + | - Coordination among generator authors of Herwig, Pythia and Sherpa (mattermost channel?) | ||
| + | - Common framework for tuning exist, but no single reproducible script: | ||
| + | - Stefan K. trying to provide this in Professor, which automatically documents what version, options, flags, etc. are being run | ||
| + | - Someone could do something similar in Apprentice | ||
| + | - Les Houches Accord on some Benchmark observable+weight sets: | ||
| + | - Observable quality & quantity requirements: | ||
| + | - No ratio Rivets if full information is available? | ||
| + | - Reduce double counting of observables: | ||
| + | - No gaps in histogram (see e.g. some ALEPH identified particle momentum distribution) | ||
| + | - Smaller bin-width is preferred | ||
| + | - Enough statistics: | ||
| + | - NOTE: If the errors of the data are to small a theory uncertainty should be provided to the tuning framework otherwise the minimization will likely over-fit this data | ||
| + | - Benchmark sets: | ||
| + | - Conservative set of selected (IRC safe?) Event Shapes: | ||
| + | - Input from perturbative experts needed | ||
| + | - Only close to the peak, where hadronization is not dominant | ||
| + | - Provide global sets from each generator to gather information | ||
| + | - Ultimate goals: | ||
| + | - Minimal set of accorded observables+weights among all generators for comparability | ||
| + | - Tune using $\chi^2$ (more strict) and $\chi^2/(1+\chi^2)$ (less strict) | ||
| + | - Maximal set of accorded observables+weights among all generators for comparability | ||
| + | - Tune using $\chi^2$ (more strict) and $\chi^2/(1+\chi^2)$ (less strict) | ||
| + | - Provide recommendations to experiments on tuning? | ||
| + | - ... | ||
| + | |||
| + | Interested: Simon, Andrzej, Stefan K., .... | ||
2023/groups/smtools/start.1686918250.txt.gz · Last modified: 2023/06/16 14:24 by josh.mcfayden
Page Tools
Except where otherwise noted, content on this wiki is licensed under the following license: Public Domain
