-
Notifications
You must be signed in to change notification settings - Fork 20
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.
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
Add new CaloCluster event class #1178
Conversation
FWIW even though many tests "failed" this appears to be due to the KS metric being too tight. Here's one random failing histogram from the PN sample, for example. (I guess since the events are generated with a different random seed they aren't identical?) My changes shouldn't affect any of these plots, so I think the failures are unrelated to the PR. |
I can't comment directly on the failing ecal plot, but the seed should be identical for the two tests. But yeah it seems strange that something like the hcal sim energy would change otherwise |
@EinarElen ok interesting. Does this pipeline test usually work? The PR literally just adds a new (unrelated) event class so I really don't think the failing plots can be related to the changes here. |
Yeah usually it works well. But yeah the failure here can't be from the changes here (unless there's something really obscure going on), so I think there's something happening on the CI side. Maybe the most recent gold histograms were mixed up with an earlier version? A very quick check for that would be to just run a small test manually once with current trunk and once with the branch. If nobody has figured it out by tomorrow, I'll give that a try :) |
Hm yeah I don't see the same pipelines run for the past 2 PRs (but maybe I'm not looking in the right place?) and the last commit from @tomeichlersmith updated the reference histograms so my bet is some kind of CI issue. |
The validation samples should be running with the same random seed and therefore should be generating identical events.1 Footnotes
|
I was able to confirm that Set UpThe
RunningGoing to do
|
It does appear to be an issue with how the golden histograms were generated. They do not match the manually generated histograms for the same tag. |
The run which produced the golden histograms appears to have cloned the correct commit of ldmx-sw (deduced from looking at the logs). https://github.com/LDMX-Software/ldmx-sw/actions/runs/5349505801 I fear that this is revealing an actual code bug but I have no idea what it could be... |
Alright, I found a difference between the environment that produced the histograms and the environment producing them now. The run producing the histograms is using
and then looking at I am going to run the inclusive sample on I forget where @EinarElen pointed this out, but I vaguely remember you mentioning that changing the compiler version may affect how the RNG behaves which seems to be the case. |
Yep, that can happen :) Great for reproducibility |
Rerunning |
I'm adding a new CaloCluster class that EcalCluster and HcalCluster can inherit from, in order to reduce code duplication and increase usability of existing tools aiming at one class or the other.
What are the issues that this addresses?
This is the first step needed to resolve #1177
Check List
I re-ran clustering locally with an EcalCluster prototype based on this base class. Results were unchanged wrt the previous implementation.
I leave updates to the submodules for future work/reviews, and will follow that progress in the issue reference above.