Add DANRA tutorial notebook with pytest-nbmake (#69)

[Sharkyii]

Describe your changes

Added hello_world_danra.ipynb, an end-to-end tutorial demonstrating neural-lam training on a small DANRA dataset (data prep → graph creation → 1-epoch CPU training → evaluation). This is taken from #202, credits: @Jayant-kernel

Enabled notebook CI using pytest-nbmake; notebooks under docs/notebooks/ now run as pytest tests, with a conftest.py fixture pre-creating danra.datastore.zarr via MDPDatastore to avoid runtime downloads, and notebook logic skipping data prep if it already exists. Dev dependencies updated with nbmake>=1.5.0 and ipykernel>=6.0.0.

Issue Link

Solves #69

Type of change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 💥 Breaking change (fix or feature that would cause existing functionality to not work as expected)
• 📖 Documentation (Addition or improvements to documentation)

Checklist before requesting a review

• My branch is up-to-date with the target branch - if not update your fork with the changes from the target branch (use pull with --rebase option if possible).
• I have performed a self-review of my code
• For any new/modified functions/classes I have added docstrings that clearly describe its purpose, expected inputs and returned values
• I have placed in-line comments to clarify the intent of any hard-to-understand passages of my code
• I have updated the README to cover introduced code changes
• I have added tests that prove my fix is effective or that my feature works
• I have given the PR a name that clearly describes the change, written in imperative form (context).
• I have requested a reviewer and an assignee (assignee is responsible for merging). This applies only if you have write access to the repo, otherwise feel free to tag a maintainer to add a reviewer and assignee.

Checklist for reviewers

Each PR comes with its own improvements and flaws. The reviewer should check the following:

• the code is readable
• the code is well tested
• the code is documented (including return types and parameters)
• the code is easy to maintain

Author checklist after completed review

• I have added a line to the CHANGELOG describing this change, in a section
  reflecting type of change (add section where missing):
  • added: when you have added new functionality
  • changed: when default behaviour of the code has been changed
  • fixes: when your contribution fixes a bug
  • maintenance: when your contribution is relates to repo maintenance, e.g. CI/CD or documentation

Checklist for assignee

• PR is up to date with the base branch
• the tests pass
• (if the PR is not just maintenance/bugfix) the PR is assigned to the next milestone. If it is not, propose it for a future milestone.
• author has added an entry to the changelog (and designated the change as added, changed, fixed or maintenance)
• Once the PR is ready to be merged, squash commits and merge the PR.

[Sharkyii]

@sadamov once free have a look here :)

[sadamov]

Okay I organized the hello_world issue and PRs:

• Add DANRA tutorial notebook with pytest-nbmake #69 now exclusively talks about DANRA -> assigned to @Sharkyii
• Add DANRA tutorial notebook with pytest-nbmake (#69) #577 is the official and only PR for the DANRA notebook
• Add COSMO example notebook for onboarding #587 is a new issue only targeting COSMO -> would like to assign to @info-gallary and @PalaniappanR02 to collaborate
• Add COSMO example notebook for onboarding (#587) #392 is the existing COSMO notebook that now closes Add COSMO example notebook for onboarding #587

If there was some oversight let me know, tried my best to look through all previous comms.

[sadamov]

Thanks for taking over here @Sharkyii. We are almost done! Please see inline suggestions for CHANGELOG.md, pyproject.toml, and the workflow, plus the notebook notes below.

Make sure sure to fully render the notebook before the next review. And also track the other PRs that implement model weight loading (fixes the workaround below) and the implementation of WMG if it lands.

Notebook — please address before merging:

Cell 17 (graph visualisation): The current cell runs plot_graph via CLI and embeds the full graph_viz.html inline. The HTML is ~300 MB (g2m: 12 716 edges, m2g: 30 720 edges serialised as inline JSON) — this makes the notebook hang on render. Replace both the CLI cell and the display cell with a single Python cell that calls the plot_graph API directly, writes the full HTML for browser use, and displays a lightweight filtered view inline (M2M edges + mesh nodes only, <1 MB):

from IPython.display import HTML
from neural_lam.config import load_config_and_datastore
from neural_lam import utils
from neural_lam.plot_graph import plot_graph as _plot_graph

config_path = "tests/datastore_examples/mdp/danra_100m_winds/config.yaml"
_, datastore = load_config_and_datastore(config_path=config_path)
xy = datastore.get_xy("state", stacked=True)
grid_pos = xy / np.max(np.abs(xy))

graph_dir = os.path.join(datastore.root_path, "graph", "1level")
hierarchical, graph_ldict = utils.load_graph(graph_dir_path=graph_dir)

fig = _plot_graph(grid_pos=grid_pos, hierarchical=hierarchical, graph_ldict=graph_ldict)
fig.write_html("graph_viz.html", include_plotlyjs="cdn")
print("Full interactive graph saved to graph_viz.html — open in a browser.")
fig.data = tuple(t for t in fig.data if t.name in {"M2M", "Mesh nodes"})
display(HTML(fig.to_html(include_plotlyjs="cdn", full_html=False)))

Cell 23 (sitecustomize workaround): sitecustomize.py only registers argparse.Namespace. PyTorch 2.6 also rejects all neural_lam.config dataclasses when loading a checkpoint with weights_only=True, causing _pickle.UnpicklingError. Extend the safe globals list:

from neural_lam.config import (
    DatastoreSelection, ManualStateFeatureWeighting, NeuralLAMConfig,
    OutputClamping, TrainingConfig, UniformFeatureWeighting,
)
torch.serialization.add_safe_globals([
    argparse.Namespace, DatastoreSelection, ManualStateFeatureWeighting,
    NeuralLAMConfig, OutputClamping, TrainingConfig, UniformFeatureWeighting,
])

Cell 24 (eval command): --processor_layers 2 is set in the training cell but absent from the eval cell. The default is 4, so eval fails with RuntimeError: Missing key(s) in state_dict. Add --processor_layers 2 to the eval command.

Cell 25 (eval output display): The cell searches for test_rmse.pdf and pred_*.png — neither matches what eval actually writes. All outputs land as PNGs in wandb/latest-run/files/media/images/: metric plots as test_rmse_*.png, example predictions as {var}_example_*.png. No forecast zarr is produced. Replace with:

img_dir = "wandb/latest-run/files/media/images"
rmse_plots = sorted(glob.glob(os.path.join(img_dir, "test_rmse_*.png")))
if rmse_plots:
    print("RMSE scorecard:", rmse_plots[0])
    display(Image(filename=rmse_plots[0]))
else:
    print("test_rmse plot not found — check eval output above.")
example_plots = sorted(glob.glob(os.path.join(img_dir, "*_example_*.png")))
if example_plots:
    n_show = min(2, len(example_plots))
    print(f"Showing {n_show} of {len(example_plots)} prediction plot(s):")
    for p in example_plots[:n_show]:
        print(" ", p)
        display(Image(filename=p))
else:
    print("No prediction plots found — check eval output above.")

[Sharkyii]

@sadamov have a look :)

[Sharkyii]

@sadamov just have a view when free