dev

Author: davidhassell

Changelog.rst

@@ -3,9 +3,9 @@ version NEXTVERSION
 
 **2024-??-??**
 
-* Allowed ``'nearest_dtos'`` 2-d regridding to work with discrete
-  sampling geometry source grids ()
-  (https://github.com/NCAS-CMS/cf-python/issues/811)
+* Allow ``'nearest_dtos'`` 2-d regridding to work with discrete
+  sampling geometry source grids
+  (https://github.com/NCAS-CMS/cf-python/issues/832)
 * New method: `cf.Field.filled`
   (https://github.com/NCAS-CMS/cf-python/issues/811)
 * New method: `cf.Field.is_discrete_axis`

cf/data/dask_regrid.py

@@ -507,7 +507,9 @@ def _regrid(
             # Note: It is much more efficient to access
             #       'weights.indptr', 'weights.indices', and
             #       'weights.data' directly, rather than iterating
-            #       over rows of 'weights' and using 'weights.getrow'.
+            #       over rows of 'weights' and using
+            #       'weights.getrow'. Also, 'np.count_nonzero' is much
+            #       faster than 'np.any' and 'np.all'.
             count_nonzero = np.count_nonzero
             indptr = weights.indptr.tolist()
             indices = weights.indices
@@ -547,7 +549,9 @@ def _regrid(
             # Note: It is much more efficient to access
             #       'weights.indptr', 'weights.indices', and
             #       'weights.data' directly, rather than iterating
-            #       over rows of 'weights' and using 'weights.getrow'.
+            #       over rows of 'weights' and using
+            #       'weights.getrow'. Also, 'np.count_nonzero' is much
+            #       faster than 'np.any' and 'np.all'.
             count_nonzero = np.count_nonzero
             where = np.where
             indptr = weights.indptr.tolist()
@@ -580,7 +584,9 @@ def _regrid(
             # Note: It is much more efficient to access
             #       'weights.indptr', 'weights.indices', and
             #       'weights.data' directly, rather than iterating
-            #       over rows of 'weights' and using 'weights.getrow'.
+            #       over rows of 'weights' and using
+            #       'weights.getrow'. Also, 'np.count_nonzero' is much
+            #       faster than 'np.any' and 'np.all'.
             count_nonzero = np.count_nonzero
             indptr = weights.indptr.tolist()
             indices = weights.indices

cf/docstring/docstring.py

@@ -161,11 +161,9 @@
                   mapped to the closest destination point. A
                   destination point can be mapped to multiple source
                   points. Some destination points may not be
-                  mapped. Each regridded value is the sum of the
-                  contributing source elements. Masked source grid
-                  cells are mapped, but do not contribute to the
-                  regridded values. Useful for binning or for
-                  categorical data.
+                  mapped. Each regridded value is the sum of its
+                  contributing source elements. Useful for binning or
+                  for categorical data.
 
                 * `None`: This is the default and can only be used
                   when *dst* is a `RegridOperator`.""",

cf/regrid/regrid.py

@@ -674,6 +674,9 @@ def regrid(
         )
 
     if return_operator:
+        # Note: The `RegridOperator.tosparse` method will also set
+        #       'dst_mask' to False for destination points with all
+        #       zero weights.
         regrid_operator.tosparse()
         return regrid_operator
 

cf/test/test_regrid_featureType.py

@@ -25,12 +25,6 @@
     except ImportError:
         pass
 
-disallowed_methods = (
-    "conservative",
-    "conservative_2nd",
-    #    "nearest_dtos",
-)
-
 methods = (
     "linear",
     "nearest_stod",
@@ -162,7 +156,7 @@ def test_Field_regrid_grid_to_featureType_3d(self):
                     y = esmpy_regrid(coord_sys, method, src, dst, **kwargs)
 
                     self.assertEqual(y.size, a.size)
-                    self.assertTrue(np.allclose(y, a, atol=4e-3, rtol=rtol))
+                    self.assertTrue(np.allclose(y, a, atol=atol, rtol=rtol))
 
                     if isinstance(a, np.ma.MaskedArray):
                         self.assertTrue((y.mask == a.mask).all())
@@ -173,6 +167,7 @@ def test_Field_regrid_grid_to_featureType_3d(self):
     def test_Field_regrid_featureType_to_grid_2d(self):
         self.assertFalse(cf.regrid_logging())
 
+        # Create some nice data
         src = self.dst_featureType
         src.del_construct("cellmethod0")
         src = src[:12]
@@ -191,6 +186,7 @@ def test_Field_regrid_featureType_to_grid_2d(self):
         # Mask some destination grid points
         dst[0, 0, 1, 2] = cf.masked
 
+        # Expected destination regridded values
         y0 = np.ma.array(
             [[0, 0, 0, 0], [0, 0, 1122, 0], [0, 1114, 0, 0], [1106, 0, 0, 0]],
             mask=[
@@ -201,13 +197,14 @@ def test_Field_regrid_featureType_to_grid_2d(self):
             ],
         )
 
-        coord_sys = "spherical"
-
         for src_masked in (False, True):
             y = y0.copy()
             if src_masked:
                 src = src.copy()
                 src[6:8] = cf.masked
+                # This following element should be smaller, because it
+                # now only has two source cells conrtibuting to it,
+                # rather than four.
                 y[3, 0] = 547
 
             # Loop over whether or not to use the destination grid
@@ -327,7 +324,7 @@ def test_Field_regrid_featureType_bad_methods(self):
         dst = self.dst_featureType.copy()
         src = self.src_grid.copy()
 
-        for method in disallowed_methods:
+        for method in ("conservative", "conservative_2nd"):
             with self.assertRaises(ValueError):
                 src.regrids(dst, method=method)