-
Notifications
You must be signed in to change notification settings - Fork 23
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
GNILC templates processing #97
Conversation
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
thanks @giuspugl
|
yeah sorry you can neglect this one . |
@giuspugl is the |
it doesn't seem so: |
thanks @giuspugl it is only a mask I think I can share it. Can you please make sure it is readable and send me the location at NERSC? |
thanks @giuspugl I have been able to reexecute your notebook, but results are a bit different. Here is the original notebook you added in this pull request: here is my clean execution, I have deleted all spectra in https://nbviewer.org/gist/zonca/feab8cf34805fddd95c2b0fb340f0c04 Some possible differences:
Plots |
thanks @zonca,
no those are the same masks as the one released in the file I think you are estimating the spectra within the same mask. I would change codecell #58 and #59 of your notebook with :
|
@giuspugl changed that, plots are still different, see https://nbviewer.org/gist/zonca/a288a1f6964f8987e286723cba63e794 |
@@ -0,0 +1,1226 @@ | |||
{ |
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.
Line #40. binning = nmt.NmtBin(nside=nside, nlb=nlb, lmax=lmax, is_Dell=False)
this might be
binning = nmt.NmtBin(nside=nside, nlb=nlbins, lmax=lmax, is_Dell=False)
Reply via ReviewNB
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.
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.
yes, I had changed already nlbins
@giuspugl I executed the notebook again, see https://nbviewer.org/gist/zonca/ff5176b4d6245ebe3ddd657706724bc3 unfortunately spectra are still different: can you please put a cleanly executed version of your notebook with no pre-existing npz spectra on Gist so I can compare cell-to-cell and find differences? |
Seems there is an inconsistency in the spectra estimated between @zonca and me . As a test case, i suggest to run this notebook |
thanks @giuspugl |
thanks @giuspugl What should we do with the dust model? You were actually getting good results with a SLURM job and we both were getting bad results in a Notebook. |
I am worried about dtypes of the arrays, I'll make some tests. |
@giuspugl it was the dtypes. Now, not sure if this was affecting your script. |
I don't think this test is very useful, the objective is to understand what is wrong in the notebook, so I'd rather directly focus on the notebook. So I am executing the notebook carefully cell by cell cross-checking all steps. @giuspugl first thing I noticed is in cell 53:
you are turning the apodized masks into binary masks, what is the reason? doesn't it mess up the spectra with NaMaster? |
ok, I commented out that line and spectra now agree with @giuspugl and seem fine, light color is input, darker color is output: |
fully executed notebook: https://nbviewer.org/gist/zonca/515355d88ca0105e36e935ea28e9930c Let's clarify #100 first, then I can go ahead and finalize the model. |
Updated executed notebook: https://nbviewer.org/gist/zonca/b0ec6cc14fa4c22ded497cc445a7ea46 This has been run at 2048, the production runs to create the templates is the same notebook just executed at 8192. See also in this pull request the notebook |
thanks a lot @zonca that was the bug!! |
squashed to 1 commit to avoid having large notebooks in the history of the repo. Only the final execution of the notebook is merged into the docs. |
Summary
This PR addresses several open issues related to the new models of thermal dust.
#85, #84 ,#83, #69
Small scales in the IQU amplitude templates
In #74 we fit separately for TT, EE and BB fit a power law spectrum in different multipole ranges. However, as we get flatter spectral indices for polarization spectra, this will yield to injecting smaller angular scales in TT whose power at given multipole is smaller than EE and BB for all
To avoid this, in this PR we force EE and BB small scales to follow the fitted TT power law. We also remark that in this way we get EB ratio closer to ~2.
Validate outputs estimating NaMaster power spectra on both input and outputs
we have expanded our validation by means of 3 figures of merit:
Moreover, we have considered 4 masks to evaluate the quality of the small scales injected for different .
Another small difference with previous analysis is that we estimated the spectra on binned equally-spaced multipoles. We choose respectively for the 4 masks.
In notebook you can find the other validation plots.
Spectral parameters with small scales
Finally, in the last few cells of the notebook I've finalized a procedure to inject smaller angular scales to the spectral parameters.
we use the Planck GNILC maps of Bd and Td.
We inject smaller angular scales to the maps by extrapolating the power law fitted from the GNILC spectral parameter maps
Smaller angular scales are modulated similarly as the intensity amplitude map
the multipoles where the fit is performed are different given the observed spectra . In any case we don't fit beyond
ell=400
, which is consistent with the TT analysis abovegiven the fact that we inject smaller angular scales with a steeper spectral index than TT
we don't expect to injecti small scale noise when rescaling at frequencies orders of magnitude lower or larger than the reference one ( 353 GHz).