Example using Numcosmo unbinned likelihood

[combet]

Add example notebook using NC unbinned likelihood on SkySim data

[combet]

@pennalima, @vitenti, I've adapted the NC script to run on SkySim data... and it does not work :( In branch issue/9/numcosmo_unbinned

• there's the script version NumCosmo_abundance_skysim.py in the top directory
• and the notebook version notebook/NumCosmo_abundance_skysim.ipynb

I haven't been able to probably instantiate the mass and redshift in the absence of errors; the syntax cluster_z = Nc.ClusterRedshiftNodist('z-min': 0.0, 'z-max': 1.0) does not work and I haven't found the proper one.

So I reverted to using the lognormal and gaussian distributions for mass and redshift respectively. Then the script runs, but the first minimization step using NLOpt gets to lower and lower values of "Omega_c", finally reaching the lower allowed bound.

Anyhow, this is were I'm at at the moment...

[vitenti]

Hello @combet , I just pushed a new commit with a working version of the NumCosmo notebook. Please, take a look.

[combet]

@vitenti - thank you for fixing this and I saw where I got things wrong (in particular the silly mistake with the amplitude conversion). This now runs fine. I will keep experimenting and maybe move to a larger sky area/catalog to see how this scales.

I will also experiment how to launch this through the job scheduler at cc-in2p3. Quick question, will esmcmc.set_nthreads enable the job to run on 4 threads in parallel?

[vitenti]

@combet - Good news! If you want to try the parallelized version you need to use:

Ncm.func_eval_set_max_threads (4)
esmcmc.set_nthreads (4)

The first call controls the number of threads available to NumCosmo as a whole, and the second the number of threads used in sampling. Please, tell me if it works properly. There is also the MPI version, but for that it's probably easier to call directly from a script and not from a notebook