PoUnce (Propagation Of UNCErtainties) is a framework fully automatized runs of non-intrusive forward UQ simulations. It is designed for efficiency on HPC clusters.
Extending the code and adding an API for your own baseline solver and cluster/scheduler is comparatively simple due to the light-weight nature and modular design of PoUnce. PoUnce therefore hopes to head-start for your own UQ implementation.
Required Python packages are given in src/requirements.txt.
Runs are configured using a YAML input file. Example input files are located in the ini folder.
For a test run, go to ini/internal_local and run
python3 ../../src/pounce.py parameter_mlmc.yml
Run progress and results should be printed to stdout, ending with a QoI table and the sentence "PoUnce Finished". Further, results should be written to the file output_double.csv.
The authors of PoUnce are:
Jakob Dürrwächter
Thomas Kuhn
Fabian Meyer
Andrea Beck
Claus-Dieter Munz
FLEXI is Copyright (C) 2022, Jakob Dürrwächter, Prof. Dr. Andrea Beck, and Prof. Dr. Claus-Dieter Munz and is released under the terms of the GNU General Public License v3.0. For the full license terms see the included license file license.
References will be added shortly. In the meantime, please cite
A. Beck, J. Dürrwächter, T. Kuhn, F. Meyer, C.-D. Munz, C. Rohde.
“hp-multilevel Monte Carlo methods for uncertainty quantification of compressible Navier–Stokes equations”.
SIAM J. Sci. Comput. (Aug. 2020).
DOI: https://doi.org/10.1137/18M1210575 \
Further documentation can be found in the doc folder. The guide can be compiled using pandoc by running
make