Muon fit improvement

[burmist-git]

The muon ring fitter needs to be improved.
The existing toy tests for muons are overly simplified.
Realistic simulations reveal significant discrepancies between the true ring and the fitted result (see the attached PNG). The fitted parameters are shown in green.

The second major component will be fitting the muon impact point while accounting for the muon’s inclination.

I stated with :
Enhance the toy model for muon ring simulation by incorporating user-defined asymmetry in both the x and y directions.
Update the muon ring fitter test to include this ring asymmetry.

[maxnoe]

Could you please provide the code that produced your images?

I can't reproduce this systematic over-estimation of the radius locally.

[burmist-git]

@maxnoe It is very simple even with current test.
~/ctapipe/src/ctapipe/image/muon/tests/test_ring_fitter.py

just use the bigger ring:

center_xs = 0.2 * u.m
center_ys = 0.2 * u.m
radius = 0.6 * u.m
width = 0.1 * u.m

But in case of non symmetric ring you will get the error even with small ring.
For non symmetric ring you will need to take this branch :muon_fit_improvement
As well you can have a plotting script : https://github.com/burmist-git/ctapipe_run_dir/blob/master/muons_ana_ctapipe.ipynb
For the plot above I need to provide you much more than just the plotting script.

[maxnoe]

Could you please provide a minimal example that re-produces this error in a self contained script or test that can just be run?

[burmist-git]

Ah sure ... You will need this branch :muon_fit_improvement
https://github.com/burmist-git/ctapipe_run_dir/blob/master/test_ring_fitter.py
I can push the test script so it will fail the tests but I did not make it so the tests are passed.

To run it you will need :

(cta-dev) > python -m pytest -s ~test_ring_fitter.py

[burmist-git]

Could you please provide a minimal example that re-produces this error in a self contained script or test that can just be run?

Did you reproduce the error ?

[maxnoe]

The test is passing for me:

❯ git status
On branch muon_fit_improvement
Your branch is up to date with 'origin/muon_fit_improvement'.

Untracked files:
  (use "git add <file>..." to include in what will be committed)
        fit_muon.py

nothing added to commit but untracked files present (use "git add" to track)
❯ pytest src/ctapipe/image/muon/tests/test_ring_fitter.py
================================================= test session starts ==================================================
platform linux -- Python 3.12.9, pytest-8.3.5, pluggy-1.5.0
Running tests in src/ctapipe/image/muon/tests/test_ring_fitter.py.

Date: 2025-04-10T15:07:36

Platform: Linux-5.15.167.4-microsoft-standard-WSL2-x86_64-with-glibc2.34

Executable: /home/maxnoe/.local/conda/envs/cta-dev/bin/python3.12

Full Python Version:
3.12.9 | packaged by conda-forge | (main, Mar  4 2025, 22:48:41) [GCC 13.3.0]

encodings: sys: utf-8, locale: UTF-8, filesystem: utf-8
byteorder: little
float info: dig: 15, mant_dig: 15

Package versions:
eventio: 1.15.0
numpy: 2.1.3
scipy: 1.15.2
astropy: 7.0.1
numba: 0.61.0


rootdir: /home/maxnoe/CTAO/ctapipe
configfile: pyproject.toml
plugins: astropy-header-0.2.2, cov-6.0.0, anyio-4.8.0, requirements-0.2.0
collected 3 items

src/ctapipe/image/muon/tests/test_ring_fitter.py ...                                                             [100%]

================================================== 3 passed in 0.55s ===================================================

[burmist-git]

Ok try now.
This ring is failing to fit

[maxnoe]

I think you can always find some specific example where a fit is not getting to the true answer within some tight tolerance.

The question I am trying to understand here is whether there is really an issue, or that is just the expected resolution of such a method.

Do you have any concrete evidence, that there is a bug? Or a proposal for an improvement?

[burmist-git]

In the current state even a very good muons (momentum > 13 GeV and theta [0,0.1 deg]) and impact point about [0-13]~m are only 30 % efficient to fit the muon in a good way. The example of the ring i show is very much simple to be fitted correctly.

[maxnoe]

@burmist-git Ok, but have you identified a concrete issue? Do you propose to change the existing methods somehow? Add a new method?

[burmist-git]

@maxnoe Yes the non ideal muons (asymmetric in terms of pixel intensity) are not fitted well. For this reason i have implemented the test with intensity asymmetry.

The plan in very simple use the first method for defining the seed parameter and then fit with second method or similar one.

[maxnoe]

@burmist-git I think you misunderstand me: I am asking if you have identified a bug. The symptom I have now understood.

[burmist-git]

Yes, the kundu_chaudhuri method is good but only in case of uniform intensities ring or part of the ring:

In reality one have always muons with some intensity non uniformity. It is function of muon inclination and muon impact point. So this is main reason why I included the intensity asymmetry into the toy simulation.

The taubin_circle_fit as well have two problems:

1. It is using as initial parameters xc = yx= 0 and R = x_max / 2. And this have to be changed for the values obtained with kundu_chaudhuri methode.
2. It fit the mask only (does not include the pixel intensities).

[burmist-git]

Some updates before end of the week :-). In addition the kundu_chaudhuri is very sensitive to noise.

So one needs to run it on the mask only. And use the obtained values as a seed for munuit fit.
Blue is true values and red is a kundu_chaudhuri estimation.

[maxnoe]

Sorry for asking again, I will try to express myself more clearly:

Do you think there there is an actual bug here, as in "A programming mistake"?

Or is it "just" that the correctly implemented current methods are not well-suited?

[burmist-git]

The ring fitting procedure has no bugs. It is working correctly mean : doing exactly what it is asked to do. With little ajustement of the logic it will do much better.

[burmist-git]

The ring fitting procedure has been updated . I added the weight into the taubin fit. And added the ring fitting combined method (kundu_chaudhuri_taubin). This newly introduced method uses the chaudhuri kundu formulas (on the mask) to get the approximate initial values of r, xc and yc and feed it to taubin circle fit. The tests with much more distorted ring have been successfully passed. Note these more distorted rings (colosseum like asymmetry) could not be fitted correctly with old taubin fit without weights.

[burmist-git]

@maxnoe @kosack @mexanick looks it is ready ?

[mexanick]

LGTM. Not sure why CI jobs status is not yet reported, from the actions page it looks like it passed without issues.

[maxnoe]

@mexanick No jobs ran for the latest commit here.

[maxnoe]

Ok, with the latest push, jobs are passing

[ctao-dpps-sonarqube]

Analysis Details

0 Issues

• 0 Bugs
• 0 Vulnerabilities
• 0 Code Smells

Coverage and Duplications

• 93.30% Coverage (93.70% Estimated after merge)
• 0.00% Duplicated Code (0.70% Estimated after merge)

Project ID: cta-observatory_ctapipe_AY52EYhuvuGcMFidNyUs

View in SonarQube

[maxnoe]

@burmist-git Thanks a lot for this contribution.

It would be nice if you could report somewhere (e.g. in an ASWG call) on the improvements on the muon analysis resulting from this change

[burmist-git]

Sure, this needs to be done. However, there are many things coming up next month, including the next sprint and the ICRC.