Fix network save/load, tz-mismatch, and node-filter bugs found in 1.4.0 beta testing

[jpalm3r]

Summary

Three bugs found by a test user evaluating the network functionality ahead of the 1.4.0 release, each fixed test-first (failing regression test committed before the fix):

• Fixes Nodes-filtered network cannot be matched — time coordinate degrades to object dtype #676 — Network._build_dataframe degraded the time index to object dtype for nodes-filtered networks, because topology-only nodes store an empty DataFrame (not None), so the is not None filter never dropped them before pd.concat. Now also filters out empty frames.
• Fixes ComparerCollection.save()/load() does not support the NODE geometry type #677 — Comparer.save()/load() didn't support the node geometry type: save() only flattened raw model data for point, and load() raised NotImplementedError for node. Both now handle node the same way they already handle point.
• Fixes ms.match() raises an unclear TypeError when observation and model result timezones differ #678 — ms.match() raised a bare pandas TypeError with no context when the observation and model result disagreed on timezone-awareness. _match_space_time now checks tz compatibility up front and raises a clear ValueError naming both sides.

Test plan

• uv run ruff check src
• uv run mypy src/ --config-file pyproject.toml
• uv run pytest --disable-warnings (full suite)
• uv run pytest src/modelskill/metrics.py --doctest-modules
• Each fix has a dedicated regression test that failed before the fix and passes after

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Fixes three regressions found during 1.4.0 beta testing in the network matching and persistence workflow, improving robustness of node-filtered networks, node-geometry save/load, and timezone mismatch error reporting in ms.match().

Changes:

• Prevent Network._build_dataframe() from concatenating empty node DataFrames (which degraded the time index to object dtype) by filtering out empty frames.
• Add an explicit timezone-awareness compatibility check in _match_space_time() that raises a clear ValueError when obs/model timezones disagree.
• Extend Comparer.save()/Comparer.load() to support gtype == "node" the same way as "point" for raw model data round-tripping, with regression tests.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
tests/test_network.py	Adds regression coverage ensuring node-filtered networks preserve a DatetimeIndex time coordinate.
tests/test_match.py	Adds regression test asserting a clear ValueError on tz-aware vs tz-naive mismatch.
tests/test_comparercollection.py	Adds a node-geometry save/load round-trip regression test for raw model data.
src/modelskill/network.py	Filters out empty per-node frames before concat to avoid time index dtype corruption.
src/modelskill/matching.py	Introduces _check_timezone_compatibility() and calls it before trimming/alignment.
src/modelskill/comparison/_comparison.py	Enables node-geometry raw model data flattening on save and reconstruction on load.

---

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

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated no new comments.

[Copilot]

Pull request overview

Copilot reviewed 18 out of 23 changed files in this pull request and generated no new comments.

[Copilot]

Pull request overview

Copilot reviewed 18 out of 23 changed files in this pull request and generated no new comments.

Suppressed comments (2)

tests/test_network.py:624

• This test name and body no longer align: it says "from_res1d" and "empty_nodes_and_reaches", but it calls Network.from_mike() with non-empty nodes and duplicates the datetime-index assertions already covered by test_nodes_filtered_network_keeps_datetime_index(). Consider removing this redundant test (empty nodes/reaches is already covered below by test_from_mike_empty_nodes_and_reaches_keeps_topology_and_empty_outputs).

def test_from_res1d_empty_nodes_and_reaches_keeps_topology_and_empty_outputs():
    path_to_file = "./tests/testdata/network.res1d"
    network = Network.from_mike(path_to_file, nodes=["108", "101"], reaches=[])

    assert isinstance(network._df.index, pd.DatetimeIndex)

src/modelskill/network.py:425

• The PR description focuses on three bug fixes (#676–#678), but this change also introduces a public API shift: Network.from_res1d is removed and replaced by Network.from_mike plus a new Network.from_epanet (with companion file handling). Please either (a) update the PR description to explicitly call out this API change and its migration path, or (b) consider adding a compatibility shim (e.g., from_res1d = from_mike) if any released version could have external callers depending on the old name.

    def from_mike(
        cls,
        res: str | Path | Res1D,
        *,
        nodes: str | list[str] | None = None,
        reaches: str | list[str] | None = None,
    ) -> Network: