Add optional YAC remapping backend with NNN/conservative support and CI coverage by rajeeja · Pull Request #1440 · UXARRAY/uxarray

rajeeja · 2026-01-10T12:25:28Z

Closes #1441, Close #897

Added a YAC-backed remapping backend so UXarray can use YAC’s interpolation machinery when requested, while remaining entirely unchanged for existing UXarray workflows. The UXarray remap API now accepts backend="yac" plus yac_method and yac_options. If YAC isn’t available, throw error.
Implemented YAC NNN (nearest‑neighbor) remap using YAC’s InterpolationStack.add_nnn, mapping UXarray grid locations (nodes/edges/faces) to YAC cloud grids (point-based) and remapping via YAC’s coupling API.
Implemented YAC conservative remap for face‑centered data only, using YAC unstructured grids with face‑node connectivity. This preserves the correct conservative semantics for cell‑centered fields.
Built a YAC-to-UXarray conversion path that:
- Converts node coordinates to radians for YAC,
- Constructs YAC UnstructuredGrid from UXarray face_node_connectivity,
- Uses YAC points at the appropriate locations (CELL for conservative, CORNER for cloud).
Added a backend parity test to compare UXarray NN and YAC NNN on a tiny deterministic 3‑node grid. This avoids tie‑breaking ambiguity and provides a direct sanity check that YAC’s nearest‑neighbor agrees with UXarray’s own implementation.

hongyuchen1030 · 2026-01-28T23:41:18Z

uxarray/remap/yac.py

+        tgt_points = self._tgt_grid.def_points(np.deg2rad(tgt_lon), np.deg2rad(tgt_lat))
+        return src_points, tgt_points
+
+    def remap(self, values: np.ndarray) -> np.ndarray:


Can you be more specific here about what's "remap"? Is it conservative or non-conservative? You can reflect it either in function name or comment

hongyuchen1030

Looks good! Just add one comment about bit more comment description. Thanks

k202077 · 2026-02-03T21:21:07Z

In your application the setup of YAC should get much easier (e.g. no setting of calendar or start/end time required) by using the (unfortunately basically undocumented) Python bindings for the yac_core library.

Here is some basic documentation:
https://dkrz-sw.gitlab-pages.dkrz.de/yac/d8/d22/core_8pyx.html

This is an example using this interface:
https://dkrz-sw.gitlab-pages.dkrz.de/yac/d8/d58/test__core_8py_source.html

You can configure YAC to only build the yac_core library (which does not depend on libfyaml). In addition libnetcdf is also optional.

Copilot

Pull request overview

Adds an optional YAC-backed remapping backend to UXarray so users can route remap operations through YAC’s interpolation machinery when requested, while keeping existing UXarray behavior as the default.

Changes:

Introduces a new YAC backend implementation for NNN (nearest-neighbor) and conservative remapping.
Extends the .remap accessor API to accept backend="yac" plus yac_method/yac_options.
Adds YAC-only tests and a dedicated GitHub Actions workflow to build YAC/YAXT from source and run those tests.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
`uxarray/remap/yac.py`	New YAC backend implementation and import/loading logic.
`uxarray/remap/accessor.py`	Adds backend/method/options parameters and routes to YAC backend when requested.
`test/test_remap_yac.py`	Adds YAC integration/parity tests (skipped when YAC unavailable).
`.github/workflows/yac-optional.yml`	New optional CI job that builds YAXT+YAC and runs YAC-specific tests.
`.github/workflows/ci.yml`	Sets `MPLBACKEND=Agg` in CI environment.

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

Copilot · 2026-03-27T16:42:04Z

uxarray/remap/yac.py

+        remapper = remappers[src_dim]
+        out_flat = np.empty((flat_src.shape[0], remapper._tgt_size), dtype=np.float64)
+        for idx in range(flat_src.shape[0]):
+            out_flat[idx] = remapper.remap(flat_src[idx])
+


The remap currently loops in Python over flat_src.shape[0] and calls into YAC once per slice. For multi-dimensional data this can be a major bottleneck. If the YAC interpolation supports batched collections, consider requesting collection_size=flat_src.shape[0] and calling the interpolation once on the full (collection_size, src_size) array (or otherwise vectorizing) to avoid per-slice overhead.

Copilot · 2026-03-27T16:42:04Z

uxarray/remap/yac.py

+def _yac_remap(source, destination_grid, remap_to: str, yac_method: str, yac_kwargs):
+    _assert_dimension(remap_to)
+    destination_dim = LABEL_TO_COORD[remap_to]
+    options = _normalize_yac_method(yac_method)
+    options.kwargs.update(yac_kwargs or {})
+    ds, is_da, name = _to_dataset(source)
+    dims_to_remap = _get_remap_dims(ds)
+    remappers: dict[str, _YacRemapper] = {}
+    remapped_vars = {}
+
+    for src_dim in dims_to_remap:
+        remappers[src_dim] = _YacRemapper(
+            ds.uxgrid,
+            destination_grid,
+            src_dim,
+            destination_dim,
+            options.method,
+            options.kwargs,
+        )


PR description says conservative remap is supported for face-centered data only, but _yac_remap/_YacRemapper don’t enforce that. As written, users can request yac_method='conservative' while remapping node/edge data, which is likely to either error deep inside YAC or produce unexpected semantics. Add an explicit validation that conservative is only used when both source and destination locations are face/cell-based (e.g. src_dim=='n_face' and destination_dim=='n_face').

Copilot · 2026-03-27T16:42:05Z

test/test_remap_yac.py

+    out = uxds["latCell"].remap.nearest_neighbor(
+        destination_grid=dest,
+        remap_to="faces",
+        backend="yac",
+        yac_method="conservative",


This test exercises yac_method='conservative' through the nearest_neighbor(...) API, which is semantically confusing (it’s not nearest-neighbor). Once the public API is clarified (e.g., a dedicated conservative(...) method or a generic .remap(...) entrypoint), update this test to use the method that matches the behavior being tested.

Suggested change

out = uxds["latCell"].remap.nearest_neighbor(

destination_grid=dest,

remap_to="faces",

backend="yac",

yac_method="conservative",

out = uxds["latCell"].remap.conservative(

destination_grid=dest,

remap_to="faces",

backend="yac",

Copilot · 2026-03-27T16:42:05Z

uxarray/remap/accessor.py

+    def __call__(
+        self,
+        *args,
+        backend: str = "uxarray",
+        yac_method: str | None = None,
+        yac_options: dict | None = None,
+        **kwargs,
+    ) -> UxDataArray | UxDataset:
        """
        Shortcut for nearest-neighbor remapping.

        Calling `.remap(...)` with no explicit method will invoke
        `nearest_neighbor(...)`.
        """
-        return self.nearest_neighbor(*args, **kwargs)
+        return self.nearest_neighbor(
+            *args,
+            backend=backend,
+            yac_method=yac_method,
+            yac_options=yac_options,
+            **kwargs,
+        )


RemapAccessor.__call__ forwards yac_method=None into nearest_neighbor(...), which overrides that method’s default of 'nnn' and will cause .remap(..., backend='yac') to raise ValueError unexpectedly. Consider defaulting yac_method to 'nnn' here when backend=='yac', or only passing yac_method through when it’s not None.

Copilot · 2026-03-27T16:42:06Z

uxarray/remap/accessor.py

+        if backend == "yac":
+            from uxarray.remap.yac import _yac_remap
+
+            yac_kwargs = yac_options or {}
+            return _yac_remap(
+                self.ux_obj, destination_grid, remap_to, yac_method, yac_kwargs
+            )
        return _nearest_neighbor_remap(self.ux_obj, destination_grid, remap_to)


backend is only checked for the special-case value 'yac'; any other unexpected value silently falls back to the UXarray implementation. To avoid typos changing behavior without warning, validate backend (e.g., allow only 'uxarray'/'yac' and raise ValueError otherwise).

Copilot · 2026-03-27T16:42:06Z

uxarray/remap/accessor.py

+            from uxarray.remap.yac import _yac_remap
+
+            yac_kwargs = yac_options or {}
+            return _yac_remap(
+                self.ux_obj, destination_grid, remap_to, yac_method, yac_kwargs


When backend=='yac', inverse_distance_weighted(...) routes to _yac_remap(...), but _yac_remap only supports 'nnn' / 'conservative' and will not perform IDW. This makes inverse_distance_weighted(backend='yac') return results that don’t match the method name/contract. Consider raising NotImplementedError/ValueError for backend='yac' here until a real YAC-IDW implementation exists (or map explicitly to the equivalent YAC method if that’s intended).

Suggested change

from uxarray.remap.yac import _yac_remap

yac_kwargs = yac_options or {}

return _yac_remap(

self.ux_obj, destination_grid, remap_to, yac_method, yac_kwargs

raise NotImplementedError(

"inverse_distance_weighted with backend='yac' is not implemented. "

"The YAC backend currently supports only 'nnn' and 'conservative' "

"methods and will not perform inverse-distance-weighted remapping. "

"Use backend='uxarray' for IDW, or choose a different remapping "

"method that is supported by YAC."

Copilot · 2026-03-27T16:42:06Z

uxarray/remap/accessor.py

+        if backend == "yac":
+            from uxarray.remap.yac import _yac_remap
+
+            yac_kwargs = yac_options or {}
+            return _yac_remap(
+                self.ux_obj, destination_grid, remap_to, yac_method, yac_kwargs
+            )


When backend=='yac', bilinear(...) routes to _yac_remap(...), but _yac_remap only supports 'nnn' / 'conservative' and will not perform bilinear interpolation. This means bilinear(backend='yac') can silently do a different method than requested. Consider rejecting backend='yac' for bilinear until a true YAC-bilinear path is implemented (or wire it up explicitly).

Copilot · 2026-03-27T16:42:06Z

uxarray/remap/yac.py

+        remapped_vars[var_name] = da_out
+
+    ds_remapped = _construct_remapped_ds(
+        source, remapped_vars, destination_grid, destination_dim


_construct_remapped_ds(...) expects remap_to to be one of 'nodes'/'edges'/'faces' (used by SpatialCoordsRemapper), but _yac_remap passes destination_dim (e.g. 'n_node'). This will raise at runtime when constructing output coords. Pass the original remap_to label into _construct_remapped_ds instead of destination_dim.

Suggested change

source, remapped_vars, destination_grid, destination_dim

source, remapped_vars, destination_grid, remap_to

o First pass at YAC - expose remap, nnn and conservative

fe66a32

rajeeja self-assigned this Jan 10, 2026

rajeeja added 4 commits January 10, 2026 12:44

o Fix actions

ca75ae4

o Add MPI flags for CI

45da06f

o Explicitly set MPI

9849d14

o Use --without-regard-for-quality flag

e88975e

rajeeja requested review from erogluorhan and hongyuchen1030 January 10, 2026 14:53

hongyuchen1030 added 2 commits January 27, 2026 10:38

Merge branch 'main' into rajeeja/yac

e84d1df

Merge branch 'main' into rajeeja/yac

ba49926

hongyuchen1030 reviewed Jan 28, 2026

View reviewed changes

rajeeja added 6 commits March 25, 2026 11:17

Merge branch 'main' into rajeeja/yac

81bd563

Merge branch 'main' into rajeeja/yac

65c765f

Switch YAC remap backend to yac.core

50498d1

Merge remote-tracking branch 'origin/main' into rajeeja/yac

d810aed

Fix YAC and Windows CI workflows

d0febae

Merge remote-tracking branch 'origin/main' into rajeeja/yac

e1e1c46

rljacob requested a review from Copilot March 27, 2026 16:37

Copilot started reviewing on behalf of rljacob March 27, 2026 16:38 View session

Copilot AI reviewed Mar 27, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add optional YAC remapping backend with NNN/conservative support and CI coverage#1440

Add optional YAC remapping backend with NNN/conservative support and CI coverage#1440
rajeeja wants to merge 13 commits intomainfrom
rajeeja/yac

rajeeja commented Jan 10, 2026 •

edited

Loading

Uh oh!

hongyuchen1030 Jan 28, 2026

Uh oh!

hongyuchen1030 left a comment •

edited

Loading

Uh oh!

k202077 commented Feb 3, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Mar 27, 2026

Uh oh!

Copilot AI Mar 27, 2026

Uh oh!

Copilot AI Mar 27, 2026

Uh oh!

Copilot AI Mar 27, 2026

Uh oh!

Copilot AI Mar 27, 2026

Uh oh!

Copilot AI Mar 27, 2026

Uh oh!

Copilot AI Mar 27, 2026

Uh oh!

Copilot AI Mar 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

-            from uxarray.remap.yac import _yac_remap
-            yac_kwargs = yac_options or {}
-            return _yac_remap(
-                self.ux_obj, destination_grid, remap_to, yac_method, yac_kwargs
+            raise NotImplementedError(
+                "inverse_distance_weighted with backend='yac' is not implemented. "
+                "The YAC backend currently supports only 'nnn' and 'conservative' "
+                "methods and will not perform inverse-distance-weighted remapping. "
+                "Use backend='uxarray' for IDW, or choose a different remapping "
+                "method that is supported by YAC."

	source, remapped_vars, destination_grid, destination_dim
	source, remapped_vars, destination_grid, remap_to

Conversation

rajeeja commented Jan 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

hongyuchen1030 Jan 28, 2026

Choose a reason for hiding this comment

Uh oh!

hongyuchen1030 left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

k202077 commented Feb 3, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

rajeeja commented Jan 10, 2026 •

edited

Loading

hongyuchen1030 left a comment •

edited

Loading