Skip to content

feat: Score monitors for Score Based Ambiguity solver#8

Open
Ragansu wants to merge 57 commits intomainfrom
score-monitor
Open

feat: Score monitors for Score Based Ambiguity solver#8
Ragansu wants to merge 57 commits intomainfrom
score-monitor

Conversation

@Ragansu
Copy link
Owner

@Ragansu Ragansu commented Aug 20, 2025

This PR adds a score monitoring functionality to the score based solver.
--- END COMMIT MESSAGE ---

This helps debug and improve designing of different scoring config and optional cuts. The program has 2 parts one is the monitoring which stores all the sub score into a structure, the second part is in the root plugin which saves the score monitor results into a root.

Ragansu and others added 10 commits August 19, 2025 18:04
…s-project#4546)

- Add  a new comparator to the CompareGeometryId which sorts the objects by time if they've the same geometry id
Adds packed columns which measurably improve the compute performance of the triplet seeding
```c++
  // packed columns for performance reasons
  XY = 1 << 14,          ///< X and Y coordinates
  ZR = 1 << 15,          ///< Z and R coordinates
  XYZ = 1 << 16,         ///< X, Y, and Z coordinates
  XYZR = 1 << 17,        ///< X, Y, Z, and R coordinates
  VarianceZR = 1 << 18,  ///< Variance in Z and R directions
```
…ject#4561)

Updates the convex polygon bounds to use `span`, share more common code via the existing base-class.
@Ragansu Ragansu changed the title Score monitor feat: Score monitors for Score Based Ambiguity solver Aug 20, 2025
@github-actions
Copy link

github-actions bot commented Aug 20, 2025

📊: Physics performance monitoring for 2500c15

Full contents

physmon summary

❗️: Downstream build failure

  • eic-shell (cc @acts-project/epic-contacts)

andiwand and others added 14 commits August 21, 2025 09:04
…s-project#4564)

- cleanup
  - `SortedInCotTheta` vs `SortedByCotTheta`
  - `const` consistency
  - `2.` vs `2`
- perf
  - this version achieves similar (within measurement error of about ~1%) performance as the original ACTS code
  - it seems to me that the performance depends on auto vectorization which appears to be very fragile
Since time has envolved, ExaTrkX seems not the correct label anymore for the GNN pipeline implementation.
…cts-project#4523)

- Add new algorithm to digitize hits in the muon spectrometer to MuonSpacePoints
*  Truth G4 hits are smeared by either the strip-pitch uncertainty or the drift-radius uncertainty extracted from ATLAS calibration data
* Duplicate hits inside the same straw volume  / on the same virtual strip are removed by dead time
* Visualization of the digitized hits
- Add MUonSpacePoint calibrator to handle the hit uncertainties & also to later calibrate the space points during the segment fitting
acts-project#4566)

It does not seem beneficial to use Eigen types for storage. For this reason I decouple the new space point EDM from Eigen. Users can still use `Eigen::Map` for high-level operations downstream.
I figured out that at least in the compile flags of the CUDA compilation units there seems to be the `-G` option present, which enables full device debug info. By specifying `-dopt=on` I hope that we only generate limited debug information, according to NVCC documentation.
I did not see a significant change in latencies, but it feels saver to have this flag on.
… the MultiWireVolumeBuilder (acts-project#4577)

This PR decouples the creation of the volume with the navigaiton policy factory creation in order to be able to attach the navigation policy factory either later on and externally to the volume or to a blueprint node
…ectly (acts-project#4578)

- Add helper functions to calculate the chi square of a space point w.r.t. a line without instantiating a CompSpacePointAuxiliaries object
…cts-project#4576)

- Add constructor which can directly takes the paremeters
- Allow that the parameterVector may larger than the needed size of 4. The first 4 components are interpreted as the line parameters then
- Replace static_cast with toUnderlying
andiwand and others added 21 commits August 28, 2025 11:20
Fixes a few cases where Acts was used instead of ACTS
This previously in correctly stated that the track state indices remain as they were, but I've changed this behavior without adjusting the comment accordingly.
Sometimes it's useful to be able to put the output files in tests in the working directory, and not clean them up. With this PR, if `ACTS_NO_TEST_TEMP` is set, the temporary directory helper will just point at the local working directory, and not delete files on destruction.
…intersection (acts-project#4551)" (acts-project#4590)

This affects ITk tracking drastically for currently unknown reasons. Let's revert this for now and follow up later
…4193)

This PR adds / fixes the SurfaceArrray Svg conversion for Gen1 and Gen3.



Co-authored-by: Paul Gessinger <1058585+paulgessinger@users.noreply.github.com>
This is not present in Athena and comes with a measurable performance penalty
This is not present in Athena and comes with a measurable performance penalty. Parallel to acts-project#4533
Makes `Acts::Surface::s_surfaceTypeNames` `constexpr`. Previously it was declared as non-`const`, which ran foul of thread safety checkers.

Details and other minor improvements:
1. changed `std::array<std::string,N>` &rarr; `std::array<std::string_view,N>`, which is smaller and won't break if we have a longer string (small string optimisation allowed with `constexpr`).
2. Move `s_surfaceTypeNames` definition to the header, next to `enum SurfaceType`.
3. Include name for `SurfaceType::Other`
4. Move out stream operator definition to the `.cpp` file so `s_surfaceTypeNames` is only instantiated once.
- use mbf smoother
- reserve space in track and track state container
- clean namespaces
- clean includes
- remove implicit track state container copy during reverse filter
- Drop redundant namespace prefixes
- Bind free functions to namespaces
After bumping from 5.3.0 to 5.3.1, the sonar action was treating the PR
number given as `"1234 "` literally, including the `"` and therefore did
not find the PR. I'm now removing the quotes and try to avoid the space
(which I think comes from a newline at the end of the PR number
artifact).
)

The gap volumes produced when resizing a cylinder r-stack was incorrect when the volume update included a z-shifted transform: the gap volume would be produced with the original transform and therefore break the stack. This PR adds a unit test for this failure mode and a fix for it.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

9 participants