v0.2.0

Highlights

This release focuses on adding flexibility for graph creation. It introduces a number of new options for connecting different graph components and arguments that give users more fine-grained control about how to build mesh graphs.

Detailed Changes

Added

• added github pull-request template to ease contribution and review process
  #18, @joeloskarsson

• Allow for specifying relative distance as rel_max_dist when connecting nodes using within_radius method.
  #19
  @joeloskarsson

• save.to_pyg can now handle any number of 1D or 2D edge or node features when
  converting pytorch-geometric Data objects to torch.Tensor objects.
  #31
  @maxiimilian

• Add containing_rectangle graph connection method for m2g edges
  #28
  @joeloskarsson

Changed

• Create different number of mesh nodes in x- and y-direction.
  #21
  @joeloskarsson

• Changed the refinement_factor argument into two: a grid_refinement_factor and a level_refinement_factor.
  #19
  @joeloskarsson

• Connect grid nodes only to the bottom level of hierarchical mesh graphs.
  #19
  @joeloskarsson

• Change default archetypes to match the graph creation from neural-lam.
  #19
  @joeloskarsson

Fixed

• Fix wrong number of mesh levels when grid is multiple of refinement factor
  #26
  @joeloskarsson

• Fix attribute keyword bug in save function
  #35
  @joeloskarsson

Maintenance

• Ensure that cell execution doesn't time out when building jupyterbook based
  documentation #25,
  @leifdenby

Upgrade Steps

Install the new version from pypi, e.g. python -m pip install weather-model-graphs==0.2.0

Links

• Full Changelog

v0.1.0

First tagged release of weather-model-graphs which includes functionality to create three graph archetypes (Keisler nearest-neighbour, GraphCast multi-range and Oskarsson hierarchical graphs) deliniating the different connectivity options, background on graph-based data-driven models, 2D plotting utilities, JupyterBook based documentation. In this version the graph assumes grid coordinates are Cartesian coordinates.