search

ndeutschmann / zunis

Normalizing flows for neural importance sampling
https://zunis.readthedocs.io
MIT License
32 stars 6 forks source link

Benchmarking example for integration with Madgraph cross-sections #4

Closed NGoetz closed 3 years ago

NGoetz commented 3 years ago

The proposed changes introduce an example for the interfacing of Zunis to MadGraph 5 to show its application on cross-sections. The execution of exec_benchmark.sh reads in user-defined paths to MadGraph and LHAPDF. MadGraph is executed on a user-defined script and outputs a Fortran-file in the integrand folder, which will be converted into a shared-object file readable by Python. Then, the benchmarking-programm itself is activated, which first determines the process-dependent number of dimensions and then prepares the integrand for the integration process, which is performed the same way as before.

A first test showed for the hyperparameter search a wide range of performance, which is many dependent on the learning rate.

Choosing only the optimal learning rate, the results are in general better.

NGoetz commented 3 years ago

The bug in TorchPS should be fixed by now.

NGoetz commented 3 years ago

I have included the requested changes in the latest commit.