During the duration of my second semester as a MSc student, I developed a MPNN ML architecture to study continuous-variable boson particles in a periodic 2D environment. The code is entirely based on JAX Python package and is was mainly trained via open-source package NetKet. The model remains quite general and can be applied to arbitrary system size and spatial dimensionality. The repository is organized as follows:
- In MPNN_model.py is the MPNN ML model,
- In MPNN_run.py is a an exmaple scipt of how to run the model with NetKet,
- In deepset_model.py is a simpler deep-NN model,
- In distances.py are defined some useful functions to compute distances amongst particles,
- In stats.py functions for post-training analysis of the system. For example, the readial correlation function is included,
- In example_runfile.run is an example of how to run everything on the EPFL HPC cluster
- In the latex folder I included my final report on the project, as well as some slides explaining the main ideas,
- In the log_files folder I included the data used to reproduce the results shown in the report,