: End-to-end numpyro-based sampling of cosmology and initial conditions on LSST footprint

LSSTDESC / bayesian-pipelines-cosmology

Bayesian Cosmological Inference from Tomographic Maps

MIT License

4 stars 1 forks source link

Open EiffL opened 11 months ago

EiffL commented 11 months ago

Here is a detailed description of what we want to do:

Technical point: Gaussian shells, jitted jax, numpyro-based
realistic LSST photo-z distribution (fixed)
realistic LSST noise and footprint (LSST Y10 - Y1)
data generated from the forward model
sampling in two steps:
- fixed latent variables, sampling cosmology
- joint sampling of cosmology and latent variables
  - nside 1024, l ~ O(2000)
  - simulated shear + IA

EiffL commented 11 months ago

Technical things that need to be solved:

[ ] Building a baseline model (@AZhou00 will open a PR)
[ ] Slow jitting of the forward (maybe we need to jit the SHT)
[ ] Adding survey realism (@nataliaporqueres will prototype a forward model with footprint+noise, etc.)
[ ] Shear normalization issue

nataliaporqueres commented 11 months ago

I pushed the function for the computing shear from convergence with the normalisation issue. The example script is testjax.py