-
Notifications
You must be signed in to change notification settings - Fork 22
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
Optimize masks #984
Optimize masks #984
Conversation
I have moved functions related to BAL and DLA masking outside the classes and numbaised them. The code seems to run fine (at least passes the test, including the new mask tests from PR #989). I also restructured the BAL masking code so that the actual computation is done inside the function |
I still need to see how much faster the code runs. I was planning to run some tests on fugu data but @alxogm maybe you can also run on Iron to see if this solves your issue or if we still need to speed up the code |
I checked the times from the speed tests here: /global/cfs/projectdirs/desi/users/iprafols/lya_working_group/picca/speed_tests Timing for the new branch
Timing for the master branch
|
I checked the final deltas in an end-to-end Fugu run and there are some small changes in some of the deltas. I tried the branch running without DLA masking and the results are equal to the master branch. I also tried the branch without BAL masking and the results are slightly different. This seems to indicate that the difference comes from changes in the DLA masking. |
not sure why the changes are not picked up by the tests |
I guess we should wait for the DLA/BAL testing branch to see which check is failing in here... |
@Waelthus @andreicuceu can you review this PR? |
agreed, order 1e-8 changes of the correlation function should not be something to be scared of. And having 30% faster calculations definitely is nice! |
@Waelthus can you approve the PR then? |
Currently, masks take too long (in some cases more than the actual continuum fitting process). This PR intends to fix this
This PR is still on-going. Some things needed before this is ready to review