Galaxy generation from COSMOS catalog

JelleAalbers commented 3 years ago

This adds:

GalaxyCatalog: can generate random dictionaries of arguments for lenstronomy.LightModel.Profiles.interpolation.Interpol, describing real galaxy images shifted to a new redshift.
COSMOSCatalog: as above, using the COSMOS/GREAT3 23.5 magnitude catalog as used in galsim: https://github.com/GalSim-developers/GalSim/wiki/RealGalaxy%20Data
Refactor the cosmology initialization to a separate function (since we'll need it for sources as well as substructure)
Add .gitignore and update requirements.txt

You can use this as follows:

import manada

cosmos_folder = '/path/to/COSMOS_23.5_training_sample'

catalog = manada.Sources.cosmos.COSMOSCatalog(cosmos_folder, cosmology='planck18')

for kwargs in catalog.iter_lightmodel_kwargs_samples(
        n_galaxies=42, z_new=4, min_mag=22, min_size_in_pixels=64):
    # ... Lenstronomy code here

A basic test for the generic GalaxyCatalog is included. However,

I'm not sure if it makes sense to test the COSMOS catalog interface, as we can't run these tests on continuous integration (without downloading and unzipping the several-GB catalog first)
I have yet to test what actually happens when feeding these kwargs to actual lenstronomy code.
... for that matter, I haven't checked the images at all, apart from that they contain finite numbers...

JelleAalbers commented 3 years ago

After taking out the spurious luminosity-distance scaling, this now produces arguments that cause lenstronomy to return the original image, in the same units/scale, in a setup with no lensing and the same pixel size. I added a notebook test_sources.ipynb as an example.

Note in particular that you have to call light_model.delete_interpol_caches() every time a new image is set. Much of my confusion was because I did not do this, and it seemed no change to the image had any effect at all...

sibirrer commented 3 years ago

After taking out the spurious luminosity-distance scaling, this now produces arguments that cause lenstronomy to return the original image, in the same units/scale, in a setup with no lensing and the same pixel size. I added a notebook test_sources.ipynb as an example.

Note in particular that you have to call light_model.delete_interpol_caches() every time a new image is set. Much of my confusion was because I did not do this, and it seemed no change to the image had any effect at all...

Sorry to hear that this caused confusion. Are you sure that if you are using the 'scale' argument within the kwargs_source to scale the image, you can not use the same interpolation multiple times at different scales, orientation and locations?

swagnercarena commented 3 years ago

@JelleAalbers and @swagnercarena both think the merge looks good. Deleting the branch

swagnercarena / paltas

Galaxy generation from COSMOS catalog #1