Graphnets for solving radiative transfer problems in stellar atmospheres
The generation of the database (in directory database) requires the
installation of the Lightweaver
package. It is recommended to create a new conda environment to run this experiment.
conda create -n graphnet python=3.8
conda activate graphnet
Clone the Lightweaver repository and install it using
python -m pip install lightweaver
Now install the following packages for running the generation of the database:
conda install -c conda-forge numpy scipy astropy mpi4py tqdm argparse
Now, you can run the generation of the database by typing:
mpiexec -n 10 python database.py --n 10000 --freq 1 --out training
mpiexec -n 10 python database.py --n 500 --freq 1 --out test
This is a computationally heavy procedure that is MPI parallelized. It will generate a few files containing temperature stratifications, column mass and optical depths, as well as departure coefficients.
Some precomputed database files obtained by randomly perturbing the FAL-C atmospheric model can be found here
The configuration of the Graphnet model is tuned with a configuration file, that
needs to be passed to the training script. An example is given by conf.dat, so that
training can be done using:
python train.py --conf=conf.dat --gpu=0
The results of the training can be checked in some cases for the test set with
python test.py
You need to install the following packages:
conda install -c conda-forge sklearn configobj
conda install pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch -c nvidia
The installation of PyTorch depends on your specific configuration. Check the webpage for more information.
This implementation of Graphnet depends on the PyTorch Geometric package. Check the webpage for installation, but here you can find the installation for PyTorch 1.9.0 with CUDA 11.1:
pip install --no-index torch-scatter -f https://pytorch-geometric.com/whl/torch-1.9.0+cu111.html
pip install torch-sparse -f https://pytorch-geometric.com/whl/torch-1.9.0+cu111.html
pip install torch-cluster -f https://pytorch-geometric.com/whl/torch-1.9.0+cu111.html
pip install torch-spline-conv -f https://pytorch-geometric.com/whl/torch-1.9.0+cu111.html
pip install torch-geometric
or do it with conda:
conda install pytorch-geometric -c rusty1s -c conda-forge