-
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
[bump minor] Cross exposure #1056
Conversation
crossexposure P1D estimate.
I am currently testing on iron: I made one run which worked with best exposure CF, and I am running separated exposure CF to compare. |
Copying slack message without plots, as it is private data: I have computed P1D on Y1 with two ways to measure the CF: |
Since the independent exposure continuum fitting case works very well, I think I will let this solution for the PR, and remove the others solutions. |
I think given the biases the "best exp" case might not be a good one to keep and having the 2 options of either using individual exposures for cont fitting (thus neglecting that the true QSO continuum should be ~ the same (although when there's long timelags between the obs it can also be the quasar changed a bit)) and co-adding (which I believe you've sped up). |
Ok, so for the last step of this PR, I will try to had an option which defines how the CF is performed, and it will also allows to separete the standard case to the cross-exposure case. |
After consultation with the desi data team, the best solution to get rid of both splitted exposures and correlated noise might be to keep only one repetition of TARGETID+NIGHT. I do not know if we add it to this PR. |
What does this mean in here? Taking only one exp or rather taking a coadd of all same-TARGETID exposures in the same night, but not coadding across nights? |
I meant taking only one exposure. I think adding a coaddition step during the night might be too complicated, and I do not think we will gain much from it. |
I guess it could be ok to only take the best of the split exposures in that case (I guess typical splits have most of the SNR in the first spectrum [conditions worsen while observing leading to an early stopping] OR have about equal T_eff when conditions were not great, but good enough for splits), but that would have less SNR than usual... Depending on how often exposures are actually split one could also think about removing splits entirely (that would keep the sample uniform, but we'd lose more spectra)... |
With this new commit, I have put all the options we wanted. |
Hi @Waelthus, Thanks a lot for all those fix ! |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
ok, I had a quick last look and think this is ok. If @iprafols is fine as well we could merge.
Basic tests are included, but end-to-end is still missing afaik. Should keep this in mind for future development.
Potentially should add the new parameters etc to the relevant drawings etc...
@@ -37,6 +37,7 @@ | |||
"rejection log file", | |||
"save format", | |||
"num processors", | |||
"delta extraction single exposure", |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
not so sure this name is really clear, potentially should add something to the parameter descriptions in the readme-files or have a more self-explanatory name. Maybe something like "spectral coadding method", but given that this is only applied for cont fitting might be misleading... @corentinravoux @iprafols any ideas?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think that as long as we explain this somewhere this name should be fine
I am fine with merging |
Large modifications of the pk1d routines, and some modifications at the delta extraction. The main point of the PR is to implement an estimator of the one-dimensional power spectrum based on cross-correlations between different exposures of the same quasar. This PR can be extended to the computation of P3D (P_cross in particular).
For P3D: Just use picca_Pk1d.py, and the new individual pk1d file contains the deltas' Fourier transforms.
For P1D cross-exposures: first use picca_delta_extraction.py with the options (use non-coadded spectra = True and keep single exposures = True) in [data], and calculate the Fourier transforms of the deltas with picca_Pk1d.py. Then, I've added a picca_Pk1D_cross_exposure.py script that estimates individual P1D cross-exposures.
The hope of this PR is to have a proper P1D cross-exposure estimator, and to start the process of implementing P3D estimators directly inside picca.