Network Support

[jpalm3r]

This pull request extends the modelskill package to support network-based model results and observations. The main focus is on integrating NetworkModelResult and NodeModelResult types throughout the matching, comparison, and data handling logic, enabling seamless workflows for network geometries alongside existing point, track, and field types.

Key changes include:

Support for Network Model Results and Observations

• Added NetworkModelResult and NodeModelResult to the public API and imports in __init__.py and model/__init__.py, and updated documentation to reflect these new types. [1] [2] [3]
• Updated type annotations and logic in matching and comparison modules to handle NodeModelResult and NetworkModelResult alongside existing model result types. [1] [2] [3] [4] [5] [6] [7] [8]

Matching and Extraction Logic

• Enhanced the match function and related helpers to support matching network observations and models, including correct extraction and alignment logic for network types. [1] [2] [3]

Data Handling and Conversion

• Introduced a _drop_scalar_coords utility to consistently remove scalar coordinates (including node) from xarray datasets when converting to pandas DataFrames, ensuring clean dataframes for all geometry types. [1] [2] [3] [4] [5]
• Modified methods for converting comparison results to observations and dataframes to handle network geometries, including new logic for the node coordinate and NodeObservation. [1] [2] [3] [4]

Attribute and Geometry Type Handling

• Improved geometry type (gtype) inference and handling throughout the codebase to support the new network type, ensuring correct behavior and error handling for all supported geometries. [1] [2]

API and Type Annotation Consistency

• Updated imports, type annotations, and public API exports to consistently include network-related types in all relevant modules. [1] [2] [3]

These changes collectively enable robust support for network-based model results and observations, aligning the handling of network geometries with existing modelskill workflows.

[Copilot]

Pull request overview

Adds initial support for network-based model results and observations in modelskill, aiming to enable matching/comparison workflows for node-indexed time series (e.g., 1D network models) alongside existing point/track workflows.

Changes:

• Introduces NetworkModelResult / NodeModelResult and NodeObservation / NetworkObservation types and wires them into match() and Comparer type annotations.
• Updates TimeSeries validation and dataframe/export helpers to allow node-based coordinates (no x/y).
• Adds a demonstration notebook for network workflows.

Reviewed changes

Copilot reviewed 14 out of 15 changed files in this pull request and generated 15 comments.

Show a summary per file

File	Description
tests/test_timeseries.py	Updates validation error-message assertions for new coordinate rules.
src/modelskill/types.py	Adds GeometryType.NETWORK and a NetworkType alias.
src/modelskill/timeseries/_timeseries.py	Allows datasets with node coordinate instead of x/y; updates x/y accessors and to_dataframe() behavior.
src/modelskill/timeseries/_point.py	Adds network-node parsing and extends coordinate handling to include network coords.
src/modelskill/timeseries/_network.py	New network parsing helper (currently has multiple runtime issues).
src/modelskill/timeseries/_coords.py	Adds NetworkCoords.
src/modelskill/timeseries/init.py	Exposes new parsing helpers.
src/modelskill/obs.py	Adds NodeObservation and NetworkObservation; extends observation() factory logic.
src/modelskill/model/network.py	Adds NetworkModelResult and NodeModelResult with node extraction and alignment logic.
src/modelskill/model/init.py	Exports the new model result classes and updates docs.
src/modelskill/matching.py	Expands NetworkObservation into NodeObservations and supports NetworkModelResult extraction.
src/modelskill/comparison/_comparison.py	Broadens raw model type hints (but several point-only assumptions remain).
src/modelskill/init.py	Exposes network model/observation types at package root (incomplete for NodeModelResult).
notebooks/Collection_systems_res1d.ipynb	Demonstrates network matching and skill computation workflow.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

@jpalm3r I've opened a new pull request, #587, to work on those changes. Once the pull request is ready, I'll request review from you.

[ecomodeller]

Review by 🤖 (Claude)

Looks good overall — the network types slot cleanly into the existing workflow, and the model/network.py vs network.py layering keeps the core package free of networkx/mikeio1d at import time.

One thing to address before merge:

CI should have a job without --group networks

All three CI workflows now install --group networks, so networkx is always available. This means CI can't detect if someone accidentally adds an unguarded import networkx in a core module (e.g. matching.py, _comparison.py).

If that happens, all CI jobs pass — but any user without pip install modelskill[networks] gets an ImportError on import modelskill.

Suggestion: Keep one test matrix entry (e.g. a single Python version in full_test.yml) that runs without --group networks. The existing test suite should pass there. The network-specific tests in test_network.py would need a skip guard:

pytest.importorskip("networkx")

or a marker at the module level.

---

Minor note: _to_observation at _comparison.py:766 checks self.gtype == "network" but GeometryType.NODE is "node". This branch is dead code — it would raise NotImplementedError for node-based Comparers if reached via cc.plot.spatial_overview(). Low priority if _to_observation is on its way out.