Skip to content

v0.4.0

Latest

Choose a tag to compare

@leifdenby leifdenby released this 28 Jul 09:23
· 5 commits to main since this release
e355f86

This release lays the foundations for introducing new mesh node layouts (through the mesh_layout argument) and for creating graphs for neural-lam (by adding save.neural_lam.torch_tensors supporting the v0.1.0 neural-lam graph format specification). In addition, this release introduces graph-filtering functionality (based on node and edge features), graph diagnostics (checking for e.g. unconnected grid nodes) and graph creation benchmarks (to support future runtime optimisation work).

Added

  • Add mesh_layout argument to mesh graph creation functions, with rectilinear as the first supported layout. Uses a two-step architecture separating coordinate creation from connectivity creation, enabling future alternative layouts (e.g. triangular). #78, @prajwal-tech07
  • Add a standalone graph consistency checking tool (wmg.diagnostics.check_graph_consistency) to ensure structural health, such as verifying all grid nodes successfully connect to the mesh #59, @AdMub
  • Add Django-style graph filtering via filter_graph, for example to select nodes by type (node__type="mesh"), edges by component (edge__component="g2m"), long edges (edge__len__gt=...), and spatial windows (node__pos__bbox=(8, 16, 8, 16)), including combined filters. #46, @leifdenby & @Joltsy10
  • Add __version__ attribute to the package init #56 @AdMub
  • Add runtime and scaling benchmarks with CLI, #117, @leifdenby
  • Add wmg.save.to_torch_tensors_on_disk to save graphs in the neural-lam tensor-on-disk format (graph storage spec v0.1.0), tested against neural-lam's own graph validator. wmg.save is restructured into a package (save.base, save.neural_lam.torch_tensors, save.neural_lam.deprecated) with all existing entrypoints re-exported unchanged. #123, @prajwal-tech07
  • Add support for writing benchmarking results to json, #140, @yuvraajnarula & @leifdenby

Deprecated

  • wmg.save.to_pyg now emits a DeprecationWarning and will not be maintained going forward; use wmg.save.to_torch_tensors_on_disk instead. #123, @prajwal-tech07

Maintenance

  • Improve isolation of README example tests by executing each code block in an isolated namespace. #65 @Shristi-Goel
  • add nb-clean to pre-commit config to ensure that committed jupyter notebooks have empty output cells (to keep notebook filesizes at minimum), #146, @leifdenby