Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Electron and proton nodes for "test" production #44

Open
jsitarek opened this issue Jun 1, 2022 · 34 comments
Open

Electron and proton nodes for "test" production #44

jsitarek opened this issue Jun 1, 2022 · 34 comments

Comments

@jsitarek
Copy link
Collaborator

jsitarek commented Jun 1, 2022

Hi @Voutsi @moralejo @rlopezcoto
As discussed today in LST1+MAGIC call we will need some electron productions for the MC test set.
I propose that we generate one low zenith node and the mid zenith nodes that are along the Crab path in mid-zenith range where we have joint data:

zen(deg) az(deg)
10.000 102.199
32.059 248.099
32.059 102.217
43.197 262.712
43.197 87.604
52.374 110.312
52.374 216.698

Energy ranges and max impacts can be taken from issue #2 (the values for gammas)
View cone is more of a problem. Shall we take the one for the protons?
@moralejo
Copy link
Collaborator

moralejo commented Jun 1, 2022 via email

@jsitarek
Copy link
Collaborator Author

jsitarek commented Jun 1, 2022 via email

@jsitarek jsitarek changed the title Electron nodes for "test" production Electron and proton nodes for "test" production Jun 8, 2022
@jsitarek
Copy link
Collaborator Author

jsitarek commented Jun 8, 2022

I realized that the shower reusage is the same (10 times) for gammas and protons in the train sample. Normally one would expect a larger number for protons as they have smaller trigger efficiency.
Are the planned shower reusage values the same for the test production?

@moralejo
Copy link
Collaborator

moralejo commented Jun 8, 2022

We have not discussed it, so I was assuming we would use the same. But feel free to argue otherwise :)

@Voutsi
Copy link
Collaborator

Voutsi commented Jun 8, 2022

In training sample, we used the same shower reusage, but we simulated more proton showers. Shall we do the same here?

@moralejo
Copy link
Collaborator

moralejo commented Jun 8, 2022

Let's see what @jsitarek thinks. With 20 reuses we would save half of the computing time. Because indeed we will need more statistics to achieve enough post-trigger (and post-analysis) events.

@jsitarek
Copy link
Collaborator Author

jsitarek commented Jun 8, 2022

I also for running with reuse of 20 for protons.

@jsitarek
Copy link
Collaborator Author

Hi @Voutsi

any update on this test production?

Cheers,
Julian

@Voutsi
Copy link
Collaborator

Voutsi commented Jun 28, 2022

Hi Julian. No. I am currently producing the > 70deg nodes for the test sample. It takes a lot of time, the curved option for Corsika is slower + we have many energetic showers. Additionally there were some problems the last 2 weekends with the shutdown of the cluster. So it takes time.

However, if the electron & proton nodes are urgent, I will stop the test dataset production once the current bunch of jobs is done and move to it. For the electrons I can produce few files to estimate the viewcone - impact limits

@jsitarek
Copy link
Collaborator Author

Hi,

@moralejo and @rlopezcoto should comment on the priorities.
I would personally say that it would be good to produce one of the zd=43 electron nodes as soon as possible - this is just one node, electrons we do not have at all in this prod2, and they should be relatively fast to produce (comparing to protons).

@Voutsi
Copy link
Collaborator

Voutsi commented Jun 28, 2022

Ok, we can have the node ready this week

@jsitarek
Copy link
Collaborator Author

great, thanks!

@moralejo
Copy link
Collaborator

moralejo commented Jun 28, 2022 via email

@Voutsi
Copy link
Collaborator

Voutsi commented Jul 1, 2022

Hi @jsitarek , you can find the electron files for 43deg at
/home/georgios.voutsinas/ws/AllSky/TestDataset/Electrons/sim_telarray
There are ~500 files of 20k showers each. I simulate with NSCAT=10 so in total there are 10^8 showers.
For the production limits: I used the energy limits from the gammas.
Scanned few values for impact area and viewcones, I think max impact = 1.2km and viewcone=8.5 are reasonable.

TrigCoresElectrons
Electrons_viewcone

@jsitarek
Copy link
Collaborator Author

jsitarek commented Jul 5, 2022

Hi @Voutsi
Thanks a lot for those files, they look fine to me. I will show some results with them tomorrow in LST1+MAGIC meeting
It would be great to have the remaining nodes from my original e-mail (in particular the ones at zd32 and 10). SInce they are so fast to generate, maybe you could run them as well.

@moralejo, what shall we do with the test sample of protons? In principle the "train" sample ones that I have now can be used for sensitivity estimations, but if you have an idea how to make the sample more complete we could generate them as well.

What would be much more useful however is helium. After the gammaness cut they are strongly eliminated, but not completely, and especially at the lowest energies there should be some background left from it. It can be also used for tuning the trigger thresholds, and I guess that if the simulations are done, they will be also used for CR spectrum study.
Since helium would take more CPU time, if it is a problem we could clip the energy range. The starting energy either way need to be increased (I would go for a factor 2 larger than protons, the classical factor is 4, but because of the fluctuations it is better to be on a safe site), and generate them only up to 10 TeV or so (the highest energies would be only interesting for the CR spectrum study)

@moralejo
Copy link
Collaborator

Hi @jsitarek , I think it is a good idea to have at some (low) zenith angle a background test sample, with protons and helium. For trigger and near-trigger studies it should be easy, because no huge statistics are needed - and one can limit the highest energies as you pointed out. For the same statistics-related reason, one can be a bit generous with the production ranges (can you make a proposal?) for trigger studies. We are also interested from the LST-1-only side to understand the trigger threshold through the hadronic shower rates.

With this low-shower-stats, trigger-oriented background sample, one can also investigate where the production ranges can be trimmed for producing a higher-stats sample which is adequate for the post-analysis studies.

@jsitarek
Copy link
Collaborator Author

Since, at least for the moment, we have only mid zenith joint Crab it would be good to have also those protons/helium at medium zenith, so I would recommend to produce them with my original Zd,Az requests (maybe skipping the one at 50deg zenith:
zen(deg) az(deg)
10.000 102.199
32.059 248.099
32.059 102.217
43.197 262.712
43.197 87.604

@moralejo
Copy link
Collaborator

Ok, good, what about the ranges? Same as in the training? Or more generous (and high-E trimmed) for the trigger- oriented sample?

@Voutsi
Copy link
Collaborator

Voutsi commented Jul 14, 2022 •
edited
Loading

Concerning the production of these files. Currently I produce the high zenith (zd=82) nodes for the test dataset. It is very slow, each node takes 3-4 days to be completed. For all the nodes remaining in (75 & 82deg) I estimate 10d - 2 weeks to be done. So shall we interrupt this production and instead start simulating the protons, He & electrons which are going to be faster?

@moralejo
Copy link
Collaborator

82 is really low... and indeed not urgent. It is fine for me to give priority to the p-He-e at the selected nodes.

@jsitarek
Copy link
Collaborator Author

Hi @moralejo sorry for late feedback, but I wanted to make some tests with the existing "train" samples.
Please find attached pdf file with the summary of those studies:
20220718_prod2_completeness.pdf

I would say that:

  • Proposed scalings of protons:
    • Energy range: (10GeV-100TeV) * cos^(-1.5)ZD
    • Impact: 1500m * cos ZD
    • Viewcone: fixed 8 deg
  • For helium:
    • Like protons, but multiply energy range by a factor of 2 (classical scaling is a factor of 4, but keeping a factor 2 of margin due to fluctuations at the lowest energies, the highest energies will be incomplete, but they are not that important for helium

Feedback is very much welcome, in particular the cos^(-1.5)ZD is scaling is just a guess, could be equally be cos^(-1)ZD

Concerning trimming at the high energy, we will use those samples also for the sensitivity calculations. We do not get into background free regime up to some 5-10 TeV, so I would say that we should have the protons at least up to 30 TeV.

@moralejo
Copy link
Collaborator

Thanks for the study, @jsitarek .
The low-E behaviour of the energy distribution as zenith increases would even suggest to keep the min E fixed... I think these must be mostly muons.
For the rest, your suggestions seem reasonable to me, you can as well check the first produced files to make sure things look ok.

@jsitarek
Copy link
Collaborator Author

even muons should have some scaling so maybe let's apply cos^-1 Zd in energy scaling, otherwise I'm a bit afraid taht we will have too many empty events.

@Voutsi Voutsi mentioned this issue Jul 22, 2022
Open
@Voutsi
Copy link
Collaborator

Voutsi commented Sep 12, 2022

Hi, the production of these nodes is completed. Please find the data at

/home/georgios.voutsinas/ws/AllSky/TestDataset

at the corresponding directories, Electrons, Protons, Helium

@jsitarek
Copy link
Collaborator Author

jsitarek commented Oct 2, 2022

Hi @Voutsi @moralejo

I had a look at the generated protons and helium and unfortunatelly only now I realized that the scaling relations that I wrote there had a stupid typo both in my final slide of the presentation and in the post above where I copy pasted from the presentation.
The scaling of the impact should not be 1500m * cos ZD, but 1500m / cos ZD. The files are indeed generated with the error: the impact is falling at higher zenith. I am really sorry for the mess, but I think the proton and helium simulations would have to be repeated.

@Voutsi
Copy link
Collaborator

Voutsi commented Oct 3, 2022

Hi @jsitarek

I also just copy the formula without really noticing the typo... Yes I agree, we need to reproduce them. I will do it asap.

@moralejo
Copy link
Collaborator

moralejo commented Oct 3, 2022 via email

@Voutsi
Copy link
Collaborator

Voutsi commented Oct 3, 2022

Hi @jsitarek

for the protons produced at zd = 10deg: Those I produced earlier in June, using energy scaling ~cos^{-2.5}, IP scaling ~cos^{-0.5}. Those are different than the proposed scaling, however for 10deg the difference is small:
IP: 1511m instead of 1523m
Emax: 103.9 TeV instead of 102.3 TeV
Also there were produced with a viewcone of 9deg instead of 8deg, which I had estimate will give a complete set (rather consevative estimation, 8deg looks also fine)

That was 20k showers with a reusage factor of 10.

Do you think we can keep them? But ok, in case you prefer to uniformise, the overhead of 2 additional nodes is not that much.

A second question: shall we go for 10k showers with a reusage of 20?

@Voutsi
Copy link
Collaborator

Voutsi commented Oct 8, 2022

Hello,

the P and He nodes have been reproduced. You can find the data at the same place, e.g. for He:

/home/georgios.voutsinas/ws/AllSky/TestDataset/Helium

@jsitarek
Copy link
Collaborator Author

jsitarek commented Oct 11, 2022 via email

@moralejo
Copy link
Collaborator

moralejo commented Oct 11, 2022 via email

@Voutsi
Copy link
Collaborator

Voutsi commented Oct 11, 2022

Hi @moralejo

I didn't use the protons from training dataset for the test sample.

Those protons (and He) were created in a dedicate production, featuring production ranges in terms of viewcone and IP in order to allow for trigger/bkg studies.

The samples were indeed created on test nodes.

Please let me know if that replies your comment

@jsitarek
Copy link
Collaborator Author

jsitarek commented Nov 3, 2022

Hi @Voutsi
can you tell me what scalings were used in the end for electrons for impact and viewcone ? I cannot figure them out from the files itself
Zd=10, impactmax = 720m, viewcone =7.5deg
Zd=32, impactmax = 850m, viewcone = 7.5 deg
Zd=43, impactmax = 1200m, viewcone = 8.5 deg
Zd=52, impactmax = 1200m, viewcone = 7.5 deg

@Voutsi
Copy link
Collaborator

Voutsi commented Nov 3, 2022

Hi @jsitarek ,

I figured out the scalings by few test jobs, by looking at the distance and off axis angles of the showers that triggered. The node zd=43 which was the first produced has rather conservative estimation, maybe wasting a bit of resources, that's why I reduced the viewcone to 7.5 deg for the other nodes. Similarly for impactmax. Please see the plots at the comment above
#44 (comment)

@gabemery gabemery mentioned this issue Jul 15, 2024
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
3 participants
@moralejo @Voutsi @jsitarek