Skip to content

A ultra-high precision package for simulating radio interferometers in python on compute clusters.

License

Notifications You must be signed in to change notification settings

RadioAstronomySoftwareGroup/pyuvsim

pyuvsim

codecov

pyuvsim is a comprehensive simulation package for radio interferometers in python.

A number of analysis tools are available to simulate the output of a radio interferometer (CASA, OSKAR, FHD, PRISim, et al), however each makes numerical approximations to enable speed ups. pyuvsim's goal is to provide a simulated instrument output which emphasizes accuracy and extensibility, and can represent the most general simulator design.

A comparison to other simulators may be found here.

pyuvsim, the Interferometer Simulator of Record

pyuvsim's primary goal is to be an interferometer simulator accurate at the level of precision necessary for 21cm cosmology science,

  1. High level of test coverage including accuracy (design goal is 97%).
  2. Testing against analytic calculations, monitored by continuous integration (see memo #XXX)
  3. Comparison with external simulations with standardized reference simulations

Usability and extensibility

A secondary goal is a community simulation environment which provides well documented and flexible code to support a diversity of use cases. Key elements of this approach include:

  1. Design for scalability across many cpus.
  2. Defining a clear, user-friendly standard for simulation design.
  3. Documentation of analytic validation and reference simulations

Physical Instrumental Effects

Each addition of new physics is validated against analytic calculations and included in a new reference simulation. Physics that have been included or are on the roadmap.

  1. Fully-polarized instrument response (complete)
  2. Polarized sources (analytic testing ~90% )
  3. Floating-point source position accuracy (complete)
  4. Full-sky field of view (complete)
  5. Exact antenna positions. (complete)
  6. Varied beam models across the array (complete, tested against analytic)
  7. Diffuse emission (complete, tested against analytic, paper in prep)
  8. Arbitrary spectrum (complete)
  9. Non-terrestrial observatories (Lunar observatory complete)
  10. Time domain sources (TODO)
  11. Ionospheric scintillation (TODO)

Citation

Please cite pyuvsim by citing our JOSS paper:

Lanman et al., (2019). pyuvsim: A comprehensive simulation package for radio interferometers in python. Journal of Open Source Software, 4(37), 1234, https://doi.org/10.21105/joss.01234

ADS Link

Installation

Simple installation via pip is available for users, developers should follow the directions under Developer Installation below.

A user-installation is achieved simply with pip install pyuvsim, or to get the bleeding-edge: pip install https://github.com/RadioAstronomySoftwareGroup/pyuvsim.

By default, mpi capabilities are not enabled -- many of the utilities provided in pyuvsim do not require it. To use the simulator within pyuvsim, you should install pyuvsim with pip install pyuvsim[sim]. Note that the pyuvsim simulator is intended to run on clusters running the linux operating system, but we test against Mac OSX as well. We test against both Open MPI and MPICH.

There are a few more optional dependencies for pyuvsim which enable some features, such as astropy_healpix to use healpix based sky catalogs or healpix beams, python-casacore for writing out measurement sets and lunarsky for simulating telescopes on the moon. If you would like these tools as well as the full simulator, install pyuvsim with pip install pyuvsim[all] (or use the [healpix], [casa] or [moon] options to only get the dependencies for each of those functionalities).

If you wish to manage dependencies manually read on.

Dependencies

If you are using conda to manage your environment, you may wish to install the following packages before installing pyuvsim:

Required:

  • astropy>=6.0
  • numpy>=1.23
  • psutil
  • pyradiosky>=1.0.1
  • python>=3.10
  • pyuvdata>=3.1.0
  • pyyaml>=5.4.1
  • scipy>=1.8
  • setuptools_scm>=8.1

Optional:

  • astropy-healpix>=1.0.2 (for using healpix based sky catalogs or beams)
  • mpi4py>=3.1.1 (for actually running simulations)
  • lunarsky>=0.2.5 (for simulating telescopes on the moon)
  • python-casacore>=3.5.2 (for writing CASA measurement sets)

Developer Installation

If you are developing pyuvsim, you will need to download and install the repository using git clone https://github.com/RadioAstronomySoftwareGroup/pyuvsim.git.

Navigate into the pyuvsim directory and run pip install . or pip install -e . for a developer install (which makes it so that you don't have to reinstall every time you change the code) Note that this will attempt to automatically install any missing dependencies. If you use anaconda or another package manager you might prefer to first install the dependencies as described in Dependencies (as well as the developer dependencies listed below).

To install without dependencies, run pip install --no-deps . (optionally with the -e flag as well).

If you want to do development on pyuvsim, in addition to the other dependencies you will need the following packages:

  • coverage
  • line-profiler
  • pre-commit
  • pytest
  • pytest-cov >= 5.0
  • pypandoc
  • sphinx

One other package, pytest-xdist, is not required, but can be used to speed up running the test suite by running tests in parallel. To use it call pytest with the -n auto option.

One way to ensure you have all the needed packages is to use the included environment.yaml file to create a new environment that will contain all the optional dependencies along with dependencies required for testing and development (conda env create -f environment.yaml). Alternatively, you can specify test, doc, or dev when installing pyuvdata (as in pip install .[dev]) to install the packages needed for testing (including coverage and linting) and documentation development; dev includes everything in test and doc.

Finally, install the pre-commit hook using pre-commit install to help prevent committing code that does not meet our style guidelines.

Inputs

A simulation requires sets of times, frequencies, source positions and brightnesses, antenna positions, and direction-dependent primary beam responses. pyuvsim specifies times, frequencies, and array configuration via a UVData object (from the pyuvdata package), source positions and brightnesses via Source objects, and primary beams either through UVBeam or AnalyticBeam objects.

  • All sources are treated as point sources, with flux specified in Stokes parameters and position in right ascension / declination in the International Celestial Reference Frame (equivalently, in J2000 epoch).
  • Primary beams are specified as full electric field components, and are interpolated in angle and frequency. This allows for an exact Jones matrix to be constructed for each desired source position.
  • Multiple beam models may be used throughout the array, allowing for more complex instrument responses to be modeled.

These input objects may be made from a data file or from a set of yaml configuration files. See Running a simulation.

Outputs

Data from a simulation run are written out to a file in any format accessible with pyuvdata. This includes:

  • uvfits
  • MIRIAD
  • uvh5

When read into a UVData object, the history string will contain information on the pyuvsim and pyuvdata versions used for that run (including the latest git hash, if available), and details on the catalog used.

Quick start guide

Example obsparam configuration files may be found in the reference_simulations directory.

  1. Install from github or pip.
  2. Run off of a parameter file with 4 MPI ranks:
mpirun -n 4 python scripts/run_param_pyuvsim.py reference_simulations/first_generation/obsparam_ref_1.1.yaml

Documentation

Documentation on how to run simulations and developer API documentation is hosted on ReadTheDocs.

Testing

pyuvsim uses the pytest package for unit testing. If you've cloned the source into a directory pyuvsim/, you may verify it as follows:

  1. Install pytest from anaconda or pip.
  2. Run the pytest from pyuvsim/
pytest

You can alternatively run python -m pytest pyuvsim or python setup.py test. You will need to have all dependencies installed.

Some tests are run in parallel using the mpi4py module. Those tests have a decorator pytest.mark.parallel(n) where n is an integer giving the number of parallel processes to run the test on. To temporarily disable parallel tests, run pytest with the option --nompi.

Where to find Support

Please feel free to submit new issues to the issue log to request new features, document new bugs, or ask questions.

How to contribute

Contributions to this package to add new features or address any of the issues in the issue log are very welcome, as are bug reports or feature requests. Please see our guide on contributing

Versioning Approach

We use a generation.major.minor format.

  • Generation - Release combining multiple new physical effects and or major computational improvements. Testing: Backed by unittests, internal model validation, and significant external comparison.
  • Major - Adds new physical effect or major computational improvement. Small number of improvements with each release. Testing: Backed by unittests, internal model validation and limited external comparison.
  • Minor - Bug fixes and small improvements not expected to change physical model and which do not include breaking API changes. Testing: Backed by unittests

We do our best to provide a significant period (usually 2 major generations) of deprecation warnings for all breaking changes to the API. We track all changes in our changelog.

Some helpful definitions

  • Physical effects: things like polarization effects, noise, ionospheric modeling, or nonterrestrial observing positions.
  • Major computational improvement: Support for new catalog types (e.g, diffuse maps), new analysis tools, changes to parallelization scheme
  • Small improvements: Better documentation or example code, outer framework redesign.

Maintainers

pyuvsim is maintained by the RASG Managers, which currently include:

  • Adam Beardsley (Arizona State University)
  • Bryna Hazelton (University of Washington)
  • Daniel Jacobs (Arizona State University)
  • Paul La Plante (University of California, Berkeley)
  • Jonathan Pober (Brown University)

Please use the channels discussed in the guide on contributing for code-related discussions. You can contact us privately if needed at [email protected].