Roundtable
Benjamin
FastML conference was good and brought new ideas and inspirations. Summary will follow.
Plan: Focus on GSF and discuss about GNN talk for CTD.
Carlo
ACTS trigger implementation. Opened a MR and different versions are running, but still need to look at the performance. Progressed on the tauon reconstruction when reconstructing with ACTS.
Attila: Do we need more than showing a failure? -> no, we will not keep TRT-related decorations until run5.
Plan: Fixing a few issues for run4 studies.
Anton
Training more ML models. Initial results with MLP network. The transfer is a bit weird, because training error goes down but validation error starts with offset and goes up.
Plan: fixing and sanity checks
Attila
xAOD has been about to be updated for already a year. Now they merged it which caused everything to fail. Plan for the next few days is to fix these issues.
dAODs cannot participate in scheme evolution. Now it is not possible to reproduce, because the production costs too much time.
Dimitra
Gen3 geometry validation. Muons and ITk look good.
Validating also similar to Gen2.
Debugging.
In blueprint the connection to outer volumes got lost.
Propagator seems to work.
Plan to finalise the validation this week.
Paul
No numerical issues anymore for navigation problems. Updated navigation solution for muon chamber discovery. There are workarounds already.
Removed some debug-messages which somehow invalidates the coverage metric. this change was triggered by a comment.
Attila: maybe ccache could be responsible.
Update way for edm4hep and podIO conversions. ddsim simhits simulation outputs should be linked correctly.
rootIO with custom classes is difficult and fragile.
Bug in geometry construction is solved now.
Andreas Salzburger
Started to pull out first PRs from python works. Mainly to fix dependency issues. Discussing with Paul how to move it in gradually.
Time smearing: error parametrisation works. Produces also some plots. But need new requirements for the tests.
Neza
Truth information match between acts standalone and athena is fine.
Plan to debug this week.
When sorting by ID-hash: Wanted to test if the model-matching/ordering works.
Noemi
Checked recent changes and reproduced performance plots from reports, but they did not get worse, which is good.
Pulled changes from PFs draft MR. Managed to compile but will investigate soon.
Alex
Investigate segfaults with pytest on LCG107 with clang19. Somehow the setup does not work anymore though, so can't get to the segfault at all. It seems there was in general a problem with dd4hep version and ODD. -> maybe replicate lcg-pip in a requirement file.
How to proceed with this meeting format? -> Yes please :)
