Simply install by doing pip install -e .
The lym1d likelihood can in principle be called directly with three dictionaries, one for cosmo parameters, one for thermo parameters (functions), and one for nuisance parameters. However, for most purposes it's likely to be easier to call it through the wrapper (located at src/lym1d_wrapper), which allows interfacing with more common tools such as cobaya or MontePython.
env MPICC=/opt/cray/pe/mpich/8.1.28/ofi/gnu/12.3/bin/mpicc conda install pip cython numpy scipy mpi4py
rm /global/homes/<firstletter_user_directory>/<your_user_directory>/.conda/envs/TEST/compiler_compat/ld
git clone [email protected]:lesgourg/class_public.git
cd class_public; make -j
cd python; python3 -m pip install .
cd ../..
git clone [email protected]:schoeneberg/lym1d.git
cd lym1d; python3 -m pip install . ; cd ..
In case of running with cobaya, the setup is as follows. First, follow the NERSC installation steps, as well as installing cobaya
pip install cobaya
Finally, the final run is as easy as
python3 run_Planque_2020.py
while updating the path inside the file to point to YOUR data (the predefined is NERSC). Alternatively
mpirun -n NP python3 run_Planque_2020.py
To install, please use the NERSC installation steps, and then use
git clone [email protected]:schoeneberg/montepython_public_lyadesi.git
In case of running with MontePython (the public_lyadesi repository), the command is simply
python3 montepython/MontePython.py run -p input/Lya_H0_eBOSS_orig_usingdesi.param -o chains/Lya_H0_eBOSS_orig_usingdesi --bestfit chains/Lya_H0_eBOSS_orig_usingdesi/Lya_H0_eBOSS_orig_usingdesi.bestfit --conf lya.conf
Here the file lya.conf needs to be created in the folder, and it should contain the usual lines, pointing to your favorite version of classy, such as for example
root = '/path/to/your/codes'
path['cosmo'] = root+'/your_favorite_class/class_public'
path['clik'] = root+'/Planck3/code/plc_3.0/plc-3.1/'
As such, it's paramount that you have pulled and manually installed (not pip-installed) a version of CLASS on your system before running. The tutorial is given on the class github.
Additional arguments are of course the usual MontePython arguments. Running with option -N 1 -f 0 should give a value of 229.647 if everything is installed correctly. The real runs can be started with larger options for -N and not including any -f. Additionally, in that case the MontePython command can be run with mpirun -np XX for multi-system MPI parallelization.
Note that the default running mode requires your data to be in the folder data/Lya_DESI in MontePython If this is not the case, you can either copy the data there, or change the arguments for the paths. In particular, there is the base_directory path.
In the case of NERSC use, this can point to /global/cfs/cdirs/desi/science/lya/y1-p1d/likelihood_files/, whereas data_path would be set to data_files/Chabanier19/. Please find also the corresponding input file with _NERSC on the corresponding MontePython directory (!).
There is no general definite way to run MontePython, and it depends a bit on the specifics of your setup. For smaller systems, running 4 MPI chains in parallel is a good idea, with up to 4 cores per chain. On larger systems, there is in principle no upper bound. Between 8 and 16 MPI chains in parallel are decent, and the number of OpenMPI threads should be at least around 4-8 per chain. An example would consist of 8 MPI chains run with 4 cores per chain. Random Access Memory requirements of class are typically not an issue, and much less MontePython, so they can be of order ~100MB per core. For the specifics of having set most parameters to nuisance mode, it is generally a good idea to take fewer OpenMPI threads and more MPI runs (e.g. 16 chains each with a single core) The cobaya wrapper interaction will be published shortly