Developing and testing integration of nessai in pycbc inference.
The base environment can be create using conda
conda env create -f base_env.yaml
This sets a minimum version for nessai that includes new features needed to support pycbc.
You then need to install a specific version of pycbc:
- pycbc: branch on fork
Note this branch is subject to change.
The tests can be run using the Makefile in the root directory, e.g.
make simple-test
Based on the examples included in pycbc inference: https://github.com/gwastro/pycbc/tree/master/examples/inference/samplers
Note: nessai cannot be used to sample 1-dimensional problems, so the test is 2-dimensional.
Test multiprocessing with nessai using the pool defined in pycbc.
multiprocessing-test just uses the default pool from pycbc.
multiprocessing-test-mpi uses MPI instead of the default pool.
Test that it's possible to resume the sampler. This test forces the sampler to finish before reaching the stopping criterion and then resumes it.
This test is based on the example BBH injection from pycbc: https://github.com/gwastro/pycbc/tree/master/examples/inference/bbh-injection
Note: with 8 processes this took 3-4 hours to complete.
The test uses the same data as the above test but a different config, this includes sampling chirp mass and mass ratio rather than component masses.