Installation¶
Most code is written in python, however timecritical parts such as descriptor and prior-mean evaluations are implemented using cython (and cymem), and need be compiled for your particular setup.
Requirements¶
Python (only tested with 3.6.3)
ASE (tested with 3.17 and newer)
Cython (tested with 0.28 and newer)
cymem (tested with 1.31.2 and newer)
mpi4py (tested with 3.0 and newer)
GPAW
Install from source¶
The code is avaliable as a tar-file gofee.tar.gz
.
After downloading the tar-file, unpack it using:
tar -zxvf gofee.tar.gz
Then run the build_code file inside the gofee-folder, to compile descriptor and prior-function, both used in the surrogate model. Do this using:
./build_code
This will compile the mentioned files for the python setup used at the time of compiling.
Finally when using the code, you need to have the gofee-folder in the PYTHONPATH. This is achieved using:
export PYTHONPATH=<path-to-folder>/gofee:$PYTHONPATH
When this is done, and assuming you have a working GPAW installation, you can run python scripts calling GOFFE using:
mpiexec --mca mpi_warn_on_fork 0 gpaw-python script_calling_GOFEE.py