search

aimalz / scippr

Supernova Cosmology Inference with Probabilistic Photometric Redshifts
4 stars 2 forks source link

Compare SCIPPR result to traditional approach #33

Open aimalz opened 7 years ago

aimalz commented 7 years ago

What is the experimental design for this test?

aimalz commented 6 years ago

The test cases we'll consider are as follows.

With a single mock dataset:

  1. select p(Ia)~1, p(z)~delta(z'), make Hubble diagram, standard cosmo analysis, scippr approach
  2. select p(Ia)>c, z~argmax[p(z)], make Hubble diagram, standard cosmo analysis, scippr approach
  3. all data, make Hubble diagram from argmax[p(t, z, mu)] only with hat(t)=1a, standard cosmo analysis, scippr approach

In parallel, with the same mock dataset:

  1. stack p(Ia) sheets of probabilistic z, mu to make approx Hubble diagram
  2. BEAMS with z=argmax[p(z)]
  3. scippr on all data

I'm starting this in a new branch now.