marEx 4.1__tar.gz → 4.1.2__tar.gz

{marex-4.1 → marex-4.1.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: marEx
-Version: 4.1
+Version: 4.1.2
 Summary: Marine Extremes Detection and Tracking
 Author-email: Aaron Wienkers <aaron@wienkers.com>
 Project-URL: Homepage, https://github.com/wienkers/marEx
@@ -21,7 +21,7 @@ Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 Requires-Dist: numpy
 Requires-Dist: xarray[complete]
-Requires-Dist: dask[complete]==2025.3.0
+Requires-Dist: dask[complete]>=2025.9
 Requires-Dist: netcdf4
 Requires-Dist: scipy
 Requires-Dist: scikit-image

{marex-4.1 → marex-4.1.2}/marEx/detect.py

@@ -819,9 +819,14 @@ def preprocess_data(
         ds["mask"] = ds.mask.compute()
         ds["dat_anomaly"] = fix_dask_tuple_array(ds.dat_anomaly)
 
-        # Patch for same dask-Zarr bug:
-        ds[coordinates["x"]] = ds[coordinates["x"]].compute()
-        ds[coordinates["y"]] = ds[coordinates["y"]].compute()
+        # Patch for same dask-Zarr bug: materialise *any* remaining Dask-backed coordinates.
+        # Auxiliary (non-index) coordinates (e.g. gridded `x`/`y` or unstructured `lon`/`lat`)
+        # become Dask-backed via `.chunk()` and otherwise retain tuple chunk references that
+        # break distributed serialisation on save. This is seen as an
+        # "AttributeError: 'tuple' object has no attribute 'size'" from dask.array.store.
+        for coord_name in list(ds.coords):
+            if is_dask_collection(ds[coord_name].data):
+                ds[coord_name] = ds[coord_name].compute()
         if "neighbours" in ds.data_vars:
             ds["neighbours"] = ds.neighbours.compute()
         if "cell_areas" in ds.data_vars:

{marex-4.1 → marex-4.1.2}/marEx/track.py

@@ -1612,7 +1612,11 @@ class tracker:
             # Generate structuring element (disk-shaped)
             y, x = np.ogrid[-R_fill : R_fill + 1, -R_fill : R_fill + 1]
             r = x**2 + y**2
-            diameter = 2 * R_fill
+            # Pad/trim width for the closing+opening morphology. Closing has reach 2*R_fill and the
+            # subsequent opening another 2*R_fill, so the total reach is 4*R_fill. Padding by only
+            # 2*R_fill (the old value) leaves an under-padded artifact in the 2R-4R band around the
+            # periodic-longitude seam. (Matches the clean_up-branch EDT fill_holes, which pads 4*R_fill.)
+            diameter = 4 * R_fill
             se_kernel = r < (R_fill**2) + 1
             mode = "wrap" if not self.regional_mode else "edge"
 

{marex-4.1 → marex-4.1.2}/marEx.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: marEx
-Version: 4.1
+Version: 4.1.2
 Summary: Marine Extremes Detection and Tracking
 Author-email: Aaron Wienkers <aaron@wienkers.com>
 Project-URL: Homepage, https://github.com/wienkers/marEx
@@ -21,7 +21,7 @@ Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 Requires-Dist: numpy
 Requires-Dist: xarray[complete]
-Requires-Dist: dask[complete]==2025.3.0
+Requires-Dist: dask[complete]>=2025.9
 Requires-Dist: netcdf4
 Requires-Dist: scipy
 Requires-Dist: scikit-image

{marex-4.1 → marex-4.1.2}/marEx.egg-info/requires.txt

@@ -1,6 +1,6 @@
 numpy
 xarray[complete]
-dask[complete]==2025.3.0
+dask[complete]>=2025.9
 netcdf4
 scipy
 scikit-image

{marex-4.1 → marex-4.1.2}/pyproject.toml

@@ -28,7 +28,7 @@ dynamic = ["version"]
 dependencies = [
     "numpy",
     "xarray[complete]",
-    "dask[complete]==2025.3.0",
+    "dask[complete]>=2025.9",
     "netcdf4",
     "scipy",
     "scikit-image",

{marex-4.1 → marex-4.1.2}/tests/test_gridded_preprocessing.py

@@ -846,3 +846,61 @@ class TestGriddedPreprocessing:
                 dimensions=self.dimensions,
                 dask_chunks=self.dask_chunks,
             )
+
+    def test_auxiliary_coords_are_materialised(self, tmp_path):
+        """Regression: auxiliary (non-index) coordinates must not remain Dask-backed.
+
+        Datasets often carry auxiliary spatial coordinates (e.g. integer grid
+        indices ``x``/``y`` alongside ``lon``/``lat``). ``.chunk()`` turns these
+        into Dask arrays, and prior to the fix they survived ``preprocess_data``
+        un-materialised. Their lazy graphs retain tuple chunk references that
+        raise ``AttributeError: 'tuple' object has no attribute 'size'`` when the
+        result is written under a distributed scheduler (e.g. ``to_netcdf`` on a
+        ``LocalCluster``). ``preprocess_data`` must therefore compute every
+        remaining Dask-backed coordinate. This invariant is scheduler- and
+        Dask-version-independent, so it is checked directly here.
+        """
+        import dask
+        import pandas as pd
+
+        times = pd.date_range("2000-01-01", "2002-12-31", freq="D")
+        nlat, nlon = 4, 4
+        da = xr.DataArray(
+            np.random.default_rng(7).standard_normal((len(times), nlat, nlon)).astype(np.float32),
+            dims=["time", "lat", "lon"],
+            coords={
+                "time": times,
+                "lat": np.arange(nlat, dtype=np.float32),
+                "lon": np.arange(nlon, dtype=np.float32),
+                # Auxiliary, non-index coordinates that become Dask-backed via .chunk()
+                "x": ("lon", np.arange(nlon, dtype=np.int64)),
+                "y": ("lat", np.arange(nlat, dtype=np.int64)),
+            },
+        ).chunk({"time": -1})
+
+        # Sanity check: the aux coords are genuinely Dask-backed on the input
+        assert dask.base.is_dask_collection(da.coords["x"].data)
+        assert dask.base.is_dask_collection(da.coords["y"].data)
+
+        result = marEx.preprocess_data(
+            da,
+            method_anomaly="fixed_baseline",
+            method_extreme="global_extreme",
+            threshold_percentile=95,
+            dimensions={"time": "time", "x": "lon", "y": "lat"},
+            dask_chunks={"time": 25},
+        )
+
+        # No coordinate in the output may remain Dask-backed (the actual fix)
+        dask_coords = [name for name in result.coords if dask.base.is_dask_collection(result[name].data)]
+        assert dask_coords == [], f"Output coordinates still Dask-backed: {dask_coords}"
+
+        # The aux coords must still be present and carry the right values
+        assert "x" in result.coords and "y" in result.coords
+        np.testing.assert_array_equal(result["x"].values, np.arange(nlon, dtype=np.int64))
+        np.testing.assert_array_equal(result["y"].values, np.arange(nlat, dtype=np.int64))
+
+        # End-to-end: the result must be writable to NetCDF
+        out_path = tmp_path / "aux_coords_roundtrip.nc"
+        result.to_netcdf(out_path)
+        assert out_path.exists()