Overview

pressio-tools includes Python bindings providing the following functionalities:

• computation of sample mesh indices for hyper-reduction via

  • leverage scores: using algorithm 5.1 of this paper

  • other methods to come soon

• SVD for shared- or distributed-memory (tall-skinny) matrix

• QR factorization of a shared- or distributed-memory (tall-skinny) matrix

pressio-tools is self-contained but mainly developed as an auxiliary tool to pressio.

Installing

We provide two separate build/installation modes:

• to work only on shared-memory data:

  • basic mode requiring only Python and a C++ compiler; intended to get you started quickly
  • see the wiki page.

• to work on potentially distributed-memory data:

  • more advanced mode relying on MPI and Trilinos to operate on potentially distributed large-scale data
  • see the wiki page.

Usage

Interested in the leverage scores-based sample mesh functionality?

• Look at the sharedmemory demo where data is read from a single ascii file

  • the matrix is stored in a single file
  • Run as:

  export PTOOLSREPO=<fullpath-to-your-pressio-tools-clone>
  export DEMODIR=${PTOOLSREPO}/demos/samplemesh_levscores_onnode_via_driver_script
  cd ${PTOOLSREPO}/driver-scripts
  python3 hypred_levscores.py --input ${DEMODIR}/input.yaml --data-dir=${DEMODIR}

• Look at the parallel demo where data is split over multiple ascii files

  • the matrix is block-row distributed, stored as "matrix.txt.3.i", where i=0,1,2 is the MPI rank
  • Run as:

  export PTOOLSREPO=<fullpath-to-your-pressio-tools-clone>
  export DEMODIR=${PTOOLSREPO}/demos/samplemesh_levscores_parallel_via_driver_script
  cd ${PTOOLSREPO}/driver-scripts
  mpirun -n 3 python3 hypred_levscores.py --input ${DEMODIR}/input.yaml --data-dir=${DEMODIR}

Interested in the SVD functionality?

• Look at the sharedmemory demo where data is read from a single ascii file

  • the matrix is stored in a single file
  • Run as:

  export PTOOLSREPO=<fullpath-to-your-pressio-tools-clone>
  export DEMODIR=<fullpath-to-pressio-tools>/demos/svd_onnode_via_driver_script
  cd ${PTOOLSREPO}/driver-scripts
  python3 computeSVD.py --input ${DEMODIR}/input.yaml --data-dir=${DEMODIR}

• Look at the parallel demo where data is split over multiple ascii files

  • the matrix is block-row distributed, stored as "matrix.txt.3.i", where i=0,1,2 is the MPI rank
  • Run as:

  export PTOOLSREPO=<fullpath-to-your-pressio-tools-clone>
  export DEMODIR=${PTOOLSREPO}/demos/svd_parallel_via_driver_script
  cd ${PTOOLSREPO}/driver-scripts
  mpirun -n 3 python3 computeSVD.py --input ${DEMODIR}/input.yaml --data-dir=${DEMODIR}

• Look at the parallel demo where the parallel SVD is called within Python

  • the matrix is block-row distributed over 3 MPI ranks
  • Run as:

  cd <your-pressio-tools-clone>/demos/svd_parallel_call_from_python
  mpirun -n 3 python3 main.py

Questions?

Find us on Slack: https://pressioteam.slack.com or open an issue on github.

License and Citation

The full license is available here.