/xBOUT

Collects BOUT++ data from parallelized simulations into xarray.

Primary LanguagePythonApache License 2.0Apache-2.0

xBOUT

Build Status codecov

Documentation: https://xbout.readthedocs.io

xBOUT provides an interface for collecting the output data from a BOUT++ simulation into an xarray dataset in an efficient and scalable way, as well as accessor methods for common BOUT++ analysis and plotting tasks.

Currently only in alpha (until 1.0 released) so please report any bugs, and feel free to raise issues asking questions or making suggestions.

Installation

xBOUT is not currently on pip or conda. Therefore to install xBOUT on your system you must first clone the repository using:

git clone git@github.com:boutproject/xBOUT.git

or

git clone https://github.com/boutproject/xBOUT.git

Once cloned navigate to the xBOUT directory and run the following command:

pip3 install --user ./

or

python3 setup.py install

You can run the tests by navigating to the /xBOUT/ directory and entering pytest.

xBOUT requires other python packages, which will be installed when you run one of the above install commands if they are not already installed on your system.

Loading your data

The function open_boutdataset() uses xarray & dask to collect BOUT++ data spread across multiple NetCDF files into one contiguous xarray dataset.

The data from a BOUT++ run can be loaded with just

bd = open_boutdataset('./run_dir*/BOUT.dmp.*.nc', inputfilepath='./BOUT.inp')

open_boutdataset() returns an instance of an xarray.Dataset which contains BOUT-specific information in the attrs, so represents a general structure for storing all of the output of a simulation, including data, input options and (soon) grid data.

BoutDataset Accessor Methods

xBOUT defines a set of accessor methods on the loaded Datasets and DataArrays, which are called by ds.bout.method().

This is where BOUT-specific data manipulation, analysis and plotting functionality is stored, for example

ds['n'].bout.animate2D(animate_over='t', x='x', y='z')

density

or

ds.bout.create_restarts(savepath='.', nxpe=4, nype=4)

Extending xBOUT for your BOUT module

The accessor classes BoutDatasetAccessor and BoutDataArrayAccessor are intended to be subclassed for specific BOUT++ modules. The subclass accessor will then inherit all the .bout accessor methods, but you will also be able to override these and define your own methods within your new accessor.

For example to add an extra method specific to the STORM BOUT++ module:

from xarray import register_dataset_accessor
from xbout.boutdataset import BoutDatasetAccessor

@register_dataset_accessor('storm')
class StormAccessor(BoutAccessor):
    def __init__(self, ds_object):
        super().__init__(ds_object)

    def special_method(self):
        print("Do something only STORM users would want to do")

ds.storm.special_method()
Out [1]: Do something only STORM users would want to do

There is included an example of a StormDataset which contains all the data from a STORM simulation, as well as extra calculated quantities which are specific to the STORM module.

Contributing

Feel free to raise issues about anything, or submit pull requests, though I would encourage you to submit an issue before writing a pull request. For a general guide on how to contribute to an open-source python project see xarray's guide for contributors.

The existing code was written using Test-Driven Development, and I would like to continue this, so please include pytest tests with any pull requests.

If you write a new accessor, then this should really live with the code for your BOUT module, but it could potentially be added as an example to this repository too.

Development plan

Things which definitely need to be included (see the 1.0 milestone):

  • More tests, both with and against the original boutdata.collect()
  • Speed test against old collect

Things which would be nice and I plan to do:

  • Infer concatenation order from global indexes (see issue)
  • Real-space coordinates
  • Variable names and units (following CF conventions)
  • Unit-aware arrays
  • Variable normalisations

Things which might require a fair amount of effort by another developer but could be very powerful:

  • Using real-space coordinates to create tokamak-shaped plots
  • Support for staggered grids using xgcm
  • Support for encoding topology using xgcm
  • API for applying BoutCore operations (hopefully using xr.apply_ufunc)