search

LSSTDESC / firecrown

DESC Cosmology Likelihood Framework
BSD 3-Clause "New" or "Revised" License
29 stars 7 forks source link

Fix crash on S8 sampling in example #319

Closed elisachisari closed 10 months ago

elisachisari commented 11 months ago

Sampling over S8 in cosmic shear example file was broken. This is a quick fix.

marcpaterno commented 11 months ago

Elisa, can you tell us what version of the CosmoSIS Standard Library you are using? This seems to be reverting to an older version of the CSL; we would prefer, if possible, to move everyone to using the newer version.

If we need to support multiple inconsistent versions of CosmoSIS and CSL (and perhaps also NumCosmo, and Cobaya), we'll need to discuss how we should do so. We had the discussion for CCL, where the conclusion was ... don't support the older versions.

marcpaterno commented 11 months ago

@joezuntz In the current version of CosmoSIS, is the sigma8_rescale module deprecated? If not, if someone both sets sigma8 and uses the sigma8_rescale module, does anything weird happen? It appears that sigma8_rescale reads sigma8_input, which is already marked as deprecated.

tilmantroester commented 11 months ago

As of v2.5 of the CSL (and v3.3.2 of cobaya), sigma8 can be used as an input to CAMB without approximations or notable performance hits, so earlier methods like sigma8_rescale or sigma8toAs aren't required any more.

marcpaterno commented 11 months ago

These changes would revert the example to use a configuration that was required by older versions of CosmoSIS. While it also works with the newer version of CosmoSIS the result is:

  1. less efficient, and
  2. does not deal correctly with non-linear power spectra coming from CAMB.

We would prefer to reject this PR (to not revert the configuration) unless there is a need to retain examples supporting the older version of CosmoSIS.

Is there such a need?

elisachisari commented 10 months ago

Tried updating the environment to work with latest CSL and the sampling over S8 works out of the box. So I'm happy to close this.