lsstdesc / augur Goto Github PK

Currently bandpower windows are missing in the SACC file. Add them so there are no ambiguities at the time of computing the likelihood.

Pulling the (as-yet-unfinished) SN likelihood from firecrown, the SN example should be very simple, and we should be able to compare with fisher from e.g. cosmosis

When I try to use firecrown with the librefactor branch. It runs up to the cosmosis stage. Commenting out the cosmosis stage, things seem to still work, however, some modifications are needed in the yaml files to deal with this (and also in analysis.py because the cosmosis key is not going to be available).

We need to start a new branch and try to make augur compatible with the latest firecrown at some point.

Currently things like number-density and bias are hardcoded in the config file. It might be great to have some script to generate these numbers.

At the moment the tests fail if TJPCov is not installed. We should add a link to installation instructions in the README.

TJPCov is in a state of mess. At the moment, we are able to make predictions for both config and Fourier space, but only errors for Fourier space based on a simple fsky-type Gaussian covariance. The right order of things is to fix TJPCov first.

It would be great to have the capability of allowing for external priors in the Fisher analysis as in section 1.2 of App. XIII in the DETF report.

Currently, because TJPCov is not compatible with pyccl v3.0+ we had to remove the support. We need to bring this code back once TJPCov is compatible with it.

This is not a time sensitive enhancement, but we could allow for an external Fisher matrix to be brought in in order to combine experiments. (Or perhaps have an example notebook of how we should do that.)

Currently the instructions don't include pyccl as a dependency. We need to add this.

Given the work just completed for SRD on the degeneracy direction for Om/w(z), this would make an example run in augur.

It looks like I have to install pandas by hand. Check where this dependency is missing.

Reproducing SRD v1 ~exactly would be a good first test of augur.
It would take:

• implement redshift-dependent scale cut
• implement SRD v1 N(z) functional form
• spend some time creating the correct yaml config
• write FOM calculator
• DEBUG!

Update the config files to prevent the "missing" pairplots key.

Implement numerical derivatives of the likelihood with respect to the different parameters in order to get the Fisher elements and biases.

As discussed. In principle, it should all work out of the box with the next firecrown and PT integration into CCL, but unlikely in reality.

I was installing issues/20 and noticed it required pyyaml. Should we add it to the dependencies?

It looks like these two tests are failing:

FAILED augur/tests/test_analyze.py::test_analyze - ImportError: cannot import name 'build_sacc_data' from 'firecrown.ccl.two_point'
FAILED augur/tests/test_generate.py::test_generate - ImportError: cannot import name 'build_sacc_data' from 'firecrown.ccl.two_point'

because 'build_sacc_data' is no longer part of two_point.py.

This will probably require a PR in firecrown and just appropriate options switch on in augur.

We have some warnings from Astropy when running the example from the SRD:

WARNING: AstropyDeprecationWarning: output_fishery1/latex_table.tex already exists. Automatically overwriting ASCII files is deprecated. Use the argument 'overwrite=True' in the future. [astropy.io.ascii.ui] WARNING:astropy:AstropyDeprecationWarning: output_fishery1/latex_table.tex already exists. Automatically overwriting ASCII files is deprecated. Use the argument 'overwrite=True' in the future.
We should fixe them to get the desired behavior from that library in the future.

At the moment if you try to run Fisher with w as a parameter it complains about going over phantom divide, because Fisher needs to do steps around fiducial w=-1. F

Current SRD yaml files assume the "Eisenstein & Hu" power spectrum although the right one to assume is "HaloFit" if one wants to reproduce the SRD plots. Te rason behind choosing EH is that when doing the MCMC, CAMB (or CLASS?) give trouble with certain wa/w0 combinations. We should check if the newer versions of CCL provide an elegant exit for this problem already.

I feel deep envy at the littel green buttons that show the status of CI and docs. Can someone put together basic pieces like this. We want flake8 tests because we fill superior to joe.

Fix the hardcoded path in the example config_test file to the SRD covariance. It is hard to do because it needs to refer to the location of the package, but it might be feasible via environment variables.

With the current version of firecrown the tests don't work. It's just that build_sacc_data has moved locations. I'm opening a PR with the fix.

Analysis functionality is now empty. At the minimum it should implement a corner plot functionality and a list of marginalized constraints.