| CosmoChord: | PolyChord + CosmoMC for cosmological parameter estimation and evidence calculation |
|---|---|
| Author: | Will Handley |
| ForkedFrom: | https://github.com/cmbant/CosmoMC |
| Homepage: | http://polychord.co.uk |
CosmoChord is a fork of CosmoMC, which adds nested sampling provided by PolyChord.
The new Python Cobaya sampling package incorporates the latest version of PolyChord, a version of CosmoMC's sampler and most other CosmoMC features, but has more general speed optimization and general support of multiple inter-dependent theory and likelihood codes. New users should probably use that.
For full CosmoMC install details see the ReadMe.
Installation procedure:
.. bash:: git clone --recursive https://github.com/williamjameshandley/CosmoChord cd CosmoChord make export OMP_NUM_THREADS=1 ./cosmomc test.ini
To run, you should add action=5 to your ini file, and include
batch3/polychord.ini. Consider modifying test.ini
If you wish to use Planck data, you should follow the CosmoMC planck instructions, and then run make clean; make after source bin/clik_profile.sh
The master branch contains latest changes to the main release version, using latest CAMB 1.x.
The planck2018 branch contains the configuration used for the final Planck 2018 analysis, with corresponding CAMB version.
The devel branch is a development branch.
You can see the key changes by running:
.. bash:: git remote add upstream https://github.com/cmbant/CosmoMC git fetch upstream git diff --stat upstream/master git diff upstream/master source
The changes to CosmoMC are minor:
- Nested sampling heavily samples the tails of the posterior. This means that there need to be more corrections for these regions that are typically unexplored by the default metropolis hastings tool. This is now implemented by separate CAMB git submodule
- You should not use openmp parallelisation, as this in inefficient when using PolyChord. Instead, you should use pure MPI parallelisation, and you may use as many cores as you have live points.