pymor/pymor

pyMOR - Model Order Reduction with Python

pyMOR - Model Order Reduction with Python

pyMOR is a software library for building model order reduction applications with the Python programming language. All algorithms in pyMOR are formulated in terms of abstract interfaces, allowing generic implementations to work with different backends, from NumPy/SciPy to external partial differential equation solver packages.

Features

• Reduced basis methods for parametric linear and non-linear problems.
• System-theoretic methods for linear time-invariant systems.
• Neural network-based methods for parametric problems.
• Proper orthogonal decomposition.
• Dynamic mode decomposition.
• Rational interpolation of data (Loewner, AAA).
• Numerical linear algebra (Gram-Schmidt, time-stepping, ...).
• Pure Python implementations of finite element and finite volume discretizations using the NumPy/SciPy scientific computing stack.

License

pyMOR is licensed under BSD-2-clause. See LICENSE.txt.

Citing

If you use pyMOR for academic work, please consider citing our publication:

R. Milk, S. Rave, F. Schindler
pyMOR - Generic Algorithms and Interfaces for Model Order Reduction
SIAM J. Sci. Comput., 38(5), pp. S194--S216, 2016

Installation via pip

We recommend installation of pyMOR in a virtual environment.

pyMOR can easily be installed with the pip command. Please note that pip versions prior to 21.1 might have problems resolving all dependencies, so running the following first is recommended:

pip install --upgrade pip

If you are not operating in a virtual environment, you can pass the optional --user argument to pip. pyMOR will then only be installed for your local user, not requiring administrator privileges.

Latest Release (without Optional Dependencies)

For an installation with minimal dependencies, run

pip install pymor

Note that most included demo scripts additionally require Qt bindings such as pyside2 to function. Therefore we recommend install pyMOR with the gui extra:

pip install pymor[gui]  # 2023.1 and later

Latest Release (with all Optional Dependencies)

The following installs the latest release of pyMOR on your system with most optional dependencies:

pip install pymor[full]

There are some optional packages not included with pymor[full] because they need additional setup on your system:

• mpi4py: support of MPI distributed models and parallelization of greedy algorithms (requires MPI development headers and a C compiler):

  pip install mpi4py
  

• Slycot: dense matrix equation solvers for system-theoretic methods and H-infinity norm calculation (requires OpenBLAS headers and a Fortran compiler):

  pip install slycot
  

Latest Development Version

To install the latest development version of pyMOR, execute

pip install 'pymor[full] @ git+https://github.com/pymor/pymor'

which requires that the git version control system is installed on your system.

Current Release Branch Version

From time to time, the main branch of pyMOR undergoes major changes and things might break (this is usually announced in our discussion forum), so you might prefer to install pyMOR from the current release branch:

pip install 'pymor[full] @ git+https://github.com/pymor/[email protected]'

Release branches will always stay stable and will only receive bugfix commits after the corresponding release has been made.

Installation via conda

We recommend installation of pyMOR in a conda environment.

pyMOR can be installed using conda by running

conda install -c conda-forge pymor

Documentation

Documentation is available online. We recommend starting with getting started, tutorials, and technical overview.

To build the documentation locally, run the following from inside the root directory of the pyMOR source tree:

make docs

This will generate HTML documentation in docs/_build/html.

External PDE Solvers

pyMOR has been designed with easy integration of external PDE solvers in mind.

We provide bindings for the following solver libraries:

• FEniCS

  MPI-compatible wrapper classes for dolfin linear algebra data structures are shipped with pyMOR (pymor.bindings.fenics). For an example see pymordemos.thermalblock, pymordemos.thermalblock_simple. It is tested using FEniCS version 2019.1.0.

• deal.II

  Python bindings and pyMOR wrapper classes can be found here.

• NGSolve

  Wrapper classes for the NGSolve finite element library are shipped with pyMOR (pymor.bindings.ngsolve). For an example see pymordemos.thermalblock_simple. It is tested using NGSolve version v6.2.2104.

A simple example for direct integration of pyMOR with a a custom solver can be found in pymordemos.minimal_cpp_demo.

An alternative approach is to import system matrices from file and use scipy.sparse-based solvers.

Environments for pyMOR Development and Tests

Please see the Developer Documentation.

Contact

Should you have any questions regarding pyMOR or wish to contribute, do not hesitate to send us an email at

[email protected]

pymor's Issues

implement DuneRBsmLinearAffinelyDecomposedOperator(OperatorInterface)

i.e. in operatos/dune.py or combine DuneRBsmLinearOperator and LinearAffinelyDecomposedOperator

host docs on readthedocs.org

http://docs.pymor.org/ is routed properly, but building docs there produces still fails.

fix toctrees warnings during documentation generation

There seems to be a clash between numpydoc and autosummary, resulting in over 8000 warnings of the form:

<autosummary>:None: WARNING: toctree contains reference to nonexisting document 'generated/pymor.discreteoperators.basic.GenericLinearOperator.apply'

add sid to Parameters

The current key generation of Parameters based on pickling is way to slow.

add abstraction layer for numpy.ndarray so that we can easily connect to other (remote) backends

revamp caching based on sid mechanism

Maybe also use a SidRegistry?

document how to use build_ext

setup.py should automatically pick up the right version

currently it's stil hardcoded

Improve the implementation of Parametric

stabilize VectorArray interface

@ftalbrecht, please have a look at the current interface.

rename discreteoperators package to operators

Think about operators, linear operators and nonlinear operators

finalize Burgers example

implement InstationaryLinearDiscretization

implement DuneRBsmReconstructor

i.e. in reconstructors/dune.py which works on a Discretization object given by dune-rb

test Operator interface

move dune bindings from playground to a more appropriate place

test VectorArray interface

add functional implementing Neumann boundary conditions to operators.cg

implement DuneRBsmLinearOperator(OperatorInterface)

i.e. in operators/dune.py which connects to an Operator object provided by dune-rb in a shared memory environment

decide how keys for caching should be generated

add a finite volume discretization

write tutorial explaining the main concepts behind pyMor

implement reduced_dune_rb

i.e. in reductors/dune.py or see if reduced_generic_rb() works (I doubt it, becuase of the GenericReconstructor)

cleanup pandoc's redmine wiki conversion

Documentation: Show which class members are inherited by which class

This seems to be numpydocs fault. Without it, some basic information is displayed.

implement with() method for immutable objects to provide copies with changed attributes

Also consider generating warnings when changing immutable objects and providing a whitelist for attributes which actually are mutable.

Remove 'Inherits' section from docstrings

Violates DRY

add a parallelization framework to pyMOR

Decide whether a grid boundary is allowed to have multiple BoundaryTypes

This is currently undefined,

'setup.py build' fails when numpy is not installed system-wide

running build_ext
skipping 'src/pymor/tools/relations.c' Cython extension (up-to-date)
building 'pymor.tools.relations' extension
gcc -pthread -fno-strict-aliasing -DNDEBUG -g -fwrapv -O2 -Wall -Wstrict-prototypes -fPIC -I/usr/include/python2.7 -c src/pymor/tools/relations.c -o build/temp.linux-x86_64-2.7/src/pymor/tools/relations.o
src/pymor/tools/relations.c:256:31: fatal error: numpy/arrayobject.h: No such file or directory
compilation terminated.
error: command 'gcc' failed with exit status 1

Make all Grids, Operators and Discretizations immutable

importing pymor causes error if cwd not inside pyMor repo

pyMor tries to determine its version using git. If cwd is not inside the pyMor repo, this causes the following error message

In [1]: import pymor
fatal: Not a git repository (or any parent up to mount parent /home)
Stopping at filesystem boundary (GIT_DISCOVERY_ACROSS_FILESYSTEM not set).
Loading pymor version (0, 0, 0, 0, '0')

Possible ways to fix this:

1. Determine the absolute path of the pyMor installation and launch git inside the correct directory
2. Show a sensible error message if this fails for some reason.

add a dg discretization

pickle/grid tests fail when run from make test

This seems to be some weird issue with module imports from nose loader vs. pymortests.base._load_all()

fixup docstrings for sphinx

Currently there's hundreds of warnings when running sphinx on the codebase:

Parameters

pymor/src/pymor/tools/relations.py:docstring of pymor.tools.relations.inverse_relation:12: > ERROR: Unexpected indentation.
pymor/src/pymor/tools/relations.py:docstring of pymor.tools.relations.inverse_relation:13: WARNING: Block quote ends without a blank line; unexpected unindent.
uni/pymor/src/pymor/tools/relations.py:docstring of pymor.tools.relations.inverse_relation:20: SEVERE: Unexpected section title.

replace "pymor" with "pyMor" in copyright notice

I would prefer

"This file is part of the pyMor project (www.pymor.org)."

anyway.

add a way to install all dependencies when working on git repo

Patch IPython to display class inheritance when viewing class documentation

let reductors take existing reduced discretizations as an argument to speed up reduction

add POD

implement StationaryLinearDuneDiscretization(DiscretizationInterface)

i.e. in discretizations/linear_dune.py which works on a Discretization object given by dune-rb

make all demos importable

implement DuneRBsmMatrix(la.Matrix)

which can work on a Matrix object provided by dune-rb in a shared memory environment

add an unstructured grid

add missing documentation

add a basic plotting library

clean up / extend wiki frontpage

"pandoc" missing in the list of requirements

After forking and cloning pyMor, I read README.txt and executed "make". Make required pandoc, but this was not in the list of requirements in README.txt

implement discretize_elliptic_dune()

i.e. in discretizers/elliptic_dune.py which works on an example provided by dune-rb