extremeflow-pysemtools 1.3.0

No project description provided

pySEMTools

A package for post-processing data obtained using a spectral-element method (SEM), on hexahedral high-order elements.

The most prominent features of the packages are the following:

• Parallel IO: A set of routines to perform distributed IO on Nek5000/Neko field files and directly keep the data in memory on NumPy arrays or PyMech data objects.
• Parallel data interfaces: A set of objects that aim to facilitate the transfer of messages among processors. Done to ease the use of MPI functions for more inexperienced users.
• Calculus: Objects to calculate the derivation and integration matrices based on the geometry, which allows to perform calculus operations on the spectral element mesh.
• Mesh connectivity and partitioning: Objects to determine the connectivity based on the geometry and mesh repartitioning tools for tasks such as global summation, among others.
• Interpolation: Routines to perform high-order interpolation from an SEM mesh into any arbitrary query point. A crucial functionality when performing post-processing.
• Reduced-order modeling: Objects to perform parallel and streaming proper orthogonal decomposition (POD).
• Data compression/streaming: Through the use of ADIOS2 [@adios2], a set of interfaces is available to perform data compression or to connect Python scripts to running simulations to perform in-situ data processing.
• Visualization: Given that the data is available in Python, visualizations can be performed from readily available packages.

Documentation is available here.

If you wish to contribute to PySEMTools, need assistance or to report a bug, please check CONTRIBUTING.md for the community guidelines on the best way to do it.

In case you find the tools useful, please cite as:

• Perez, A., Toosi, S., Olsen, T.F., Markidis, S., Schlatter, P., 2025. Pysemtools: A library for post-processing hexahedral spectral element data. https://doi.org/10.48550/arXiv.2504.12301

The work was partially funded by the “Adaptive multi-tier intelligent data manager for Exascale (ADMIRE)” project, which is funded by the European Union’s Horizon 2020 JTI-EuroHPC research and innovation program under grant Agreement number: 956748.

Installation

There are multiple ways to install PySEMTools which are described below in more detail. For a quick-start, you can use:

pip install extremeflow-pysemtools[all]

If you are new to the package, it is probably better to get all the examples and scripts by cloning the repository instead:

# Install mpi4py (Assuming your mpi wrapper is mpicc)
env MPICC=$(which mpicc) python -m pip install --no-cache-dir mpi4py

# Install pytorch (Assuming you want PyTorch on CPUs)
python3 -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu

# Install PySEMTools and all dependencies
git clone https://github.com/ExtremeFLOW/pySEMTools.git
cd pySEMTools/
pip install --editable .[all]

For minimal functionality

To avoid cluttering clusters with many modules, the following instructions install the minimum working version of pySEMTools. This allows us to read files and perform operations with numpy in parallel.

For developers,

the easiest way to install and contribute changes is by cloning the repository:

git clone https://github.com/ExtremeFLOW/pySEMTools.git
cd pySEMTools/
pip install --editable .

Note that the --editable flag is optional, and will allow changes in the code of the package to be used directly without reinstalling.

For users,

the option to install from PyPI is available, which allows to use:

pip install extremeflow-pysemtools

For full functionality

If the objective is to be able to run all examples and tests available in the package, then more optional dependencies are needed. In this instance, the installation instruction must include the "[all]" argument, i.e.:

pip install --editable .[all]

or

pip install extremeflow-pysemtools[all]

Dependencies of note

The 3rd_party/ folder in the root of the repository contains some scripts to install their dependencies, but we can not ensure it will work in every computer. Please refer to specific library instructions or to system administrators for help when installing them.

mpi4py

mpi4py is needed even when running in serial, as the library is built with communication in mind. It can typically be installed with:

pip install mpi4py

In some instances, such as in supercomputers, it is typically necessary that the mpi of the system is used. If mpi4py is not available as a module, we have found (so far) that installing it as follows works:

export MPICC=$(which cc)
pip install mpi4py --no-cache-dir

where CC should be replaced by the correct C wrappers of the system (In a workstation you would probably need mpicc or so). It is always a good idea to contact support or check the specific documentation if things do not work.

Based on our experience on some cray systems such as Dardel and Lumi, a Wiki was made available here to help setting up on clusters.

Optional

ADIOS2

Some functionalities such as data streaming require the use of adios2. You can check how the installation is performed here.

We have also made a Wiki page that shows how we have been able to install adios2 in the supercomputers that we generally use.

An example on how adios2 can be used in conjunction with a simulation to perform, for example, insitu proper orthogonal decomposition, can be found here.

This dependency is not installed automatically with PySEMTools, since it requires the installation of the library itself with the python bindings activated.

PyTorch

Some classes are compatible with the pytorch module in case you have GPUs and want to use them in the process. We note that we only use pytorch optionally. There are versions that work exclusively with numpy on CPUs so pytorch can be avoided.

To install pytorch, you can check here. A simple installation for CUDA v12.1 on linux would look like this (following the instructions from the link):

pip3 install torch torchvision torchaudio

The process of installing pytorch in supercomputers is more intricate. In this case it is best to use the documentation of the specific cluster or contact support.

Once pytorch is available, many of the PySEMTools can be used on GPUs. We prepared examples and a Wiki page that showcases how devices can be exploited.

Catalyst2

Catalyst is a tool generally used for in-situ data visualization. It has interfaces with adios2 already, but we have also added classes that work directly with the types in PySEMTOOls.

This might be useful if you are interested in performing other insitu processing, but also want to get images.

An example on how to run it can be found here.

To perform visualizations in this way, you need to install catalyst and then install paraview with the option to use a catalyst implementation on pointing to your catalyst installation.

This dependency is not installed automatically with PySEMTools, since it requires the installation of the library itself with the python bindings activated.

HDF5

h5py is currently listed among the dependencies of the package, and it will be installed if you do not have it already. Note, however, that this installation from PyPI will most likely only work in serial. To use the parallel functionalities for the module, you will likely need to build HDF5 from source with MPI enabled, and then install h5py from this build.

Use

To get an idea on how the codes are used, feel free to check the examples we have provided. Please note that most of the routines included here work in parallel. In fact, python scripts are encouraged rather than notebooks to take advantage of this capability.

Tests

You can use the provided tests to check if your installation is complete (Not all functionalities are currently tested but more to come).

The tests rely on pytest. To install it in your pip environment simply execute pip install pytest.

Tests are performed for more functionalities than those needed to use PySEMTools in its minimal version. To run them, make sure that you use the [all] or [test] argument when installing the package to get all the dependencies (this will also install pytest).

To run the tests, execute the pytest tests/ command from the root directory of the repository. As an example, the following chain of commands will allow you to run the tests from a fresh python environment:

# Install mpi4py (Assuming your mpi wrapper is mpicc)
env MPICC=$(which mpicc) python -m pip install --no-cache-dir mpi4py

# Install pytorch (Assuming you want PyTorch on CPUs)
python3 -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu

# Install PySEMTools and all dependencies
git clone https://github.com/ExtremeFLOW/pySEMTools.git
cd pySEMTools/
pip install --editable .[test]

# Run tests
pytest tests/