Stabilize muon intensity fit #2425
Draft
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
If this fixes the numerical issue that was the root cause, why not? |
maxnoe
reviewed
Oct 27, 2023
ctapipe/image/pixel_likelihood.py
Outdated
| return np.sum(neg_log_l) | ||
| # neg_log_l provides the variable term, add constants here to only | ||
| # compute them once | ||
| return 0.5 * (len(neg_log_l) * np.log(2 * np.pi) + np.sum(neg_log_l)) |
There was a problem hiding this comment.
log(2π) could be a global constant.
There was a problem hiding this comment.
Also, do we gain something by putting an njit on this?
There was a problem hiding this comment.
I've tried, this actually resulted in a slower computation time in my tests. Probably due this function only using basic numpy functions and thus already being vectorized,
|
Docs builds failure is fixed in main, please rebase |
RuneDominik
force-pushed
the
StabilizeMuonIntensityFit
branch
from
3aba5fd
to
0ee20bb
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
See #2415
The cause of this issue might be that the likelihood values grow quite large and thus minuit does not converge properly as the values leave the [0, 1]-range by several orders of magnitude.
@maxnoe suggested to divide the likelihood used by the dof to lower the value. The likelihood is constructed to asymptotically resemble a chi2 so this changes the approach to chi2 per dof which does not change the minimum.
This does, ofc., not make the test more robust in any way.