nGC covariance #114
nGC covariance #114
Conversation
|
The tests are failing as they require a recently-updated version of CCL (LSSTDESC/CCL#1216). This PR implemented a quicker version of calculating trispectrum terms, utilized here in the calculation of the covariance. |
|
@damonge Could you bump the CCL version so we have the latest changes available? |
|
|
||
| self.cNG_conf = self.config.get("cNG", {}) | ||
|
|
||
| self.HOD_dict = { |
There was a problem hiding this comment.
We need to homogenize this with all the other halo model things in the SSC. We can do it in another pull request
…1h term regardless of whether the tracer is a numbercounts tracer or not
| @@ -15,6 +15,8 @@ def covariance_from_name(name): | |||
| from .covariance_fourier_gaussian_nmt import FourierGaussianNmt as Cov | |||
| elif name == "FourierSSCHaloModel": | |||
| from .covariance_fourier_ssc import FourierSSCHaloModel as Cov | |||
| elif name == "FouriercNGHaloModel": | |||
There was a problem hiding this comment.
You need to add also the fsky case
| cosmo: 'set' | ||
|
|
||
| # Setting mask OR fsky approximation | ||
| mask_file: |
There was a problem hiding this comment.
masks should not be required when using fsky. Remove them and check that nothing breaks (I think we already tested this for the SSC)
| a_pivot: 1.0 | ||
| ns_independent: False | ||
|
|
||
| GaussianFsky: |
There was a problem hiding this comment.
Since now more components apart from the Gaussian part will be using fsky, could you move this to the tjpcov section? i.e.
tjpcov:
fsky: 0.05
(and update the other parts of the code correspondingly)
There was a problem hiding this comment.
The only difference from the other test file is the way you compute the fsky. Cannot we merge both files so we don't need to maintain two files?
There was a problem hiding this comment.
if it is too cumbersome, i'm fine to leave it as is for now
There was a problem hiding this comment.
I suppose we could and parameterize over where to find fsky along with the bin combinations. There would be a bit of work to clean up each test combination (for gaussian, cng, and ssc) and we would need to make two sets of functions to load in the two different yamls and such. I think this format if fine for the scope of the PR and we can clean this up in a subsequent one. What do you think?
There was a problem hiding this comment.
I was thinking to merge it only for the cNG but I agree we can leave it as is for now.
There was a problem hiding this comment.
I was thinking to merge it only for the cNG but I agree we can leave it as is for now.
|
Ok. The tests pass locally so I think we can merge. I have commented the slow ones for now, until we speed the calculation of the trispectra up (in CCL). Paul's measurements: |
Implementation of the connected non-Gaussian covariance. At the moment, the trispectrum is computed for matter with a NFW halo profile and multiplied by the galaxy biases if clustering needed. This is probably wrong for the 1h term. More work needed here. At the moment, a warning is raised when using galaxy clustering.