Perf: Use ravel, not flatten, for numpy to pyarrow (#512)

See
https://stackoverflow.com/questions/28930465/what-is-the-difference-between-flatten-and-ravel-functions-in-numpy

Closes #509.

lonboard/_geoarrow/_duckdb.py

@@ -208,7 +208,7 @@ def _convert_box2d_to_geoarrow_polygon_array(
     geom_offsets = np.arange(0, len(ring_offsets), dtype=np.int32)
 
     # Construct the final PolygonArray
-    coords = pa.FixedSizeListArray.from_arrays(coords.flatten("C"), 2)
+    coords = pa.FixedSizeListArray.from_arrays(coords.ravel("C"), 2)
     ring_array = pa.ListArray.from_arrays(ring_offsets, coords)
     polygon_array = pa.ListArray.from_arrays(geom_offsets, ring_array)
     return polygon_array

lonboard/_geoarrow/extension_types.py

@@ -305,7 +305,7 @@ def construct_geometry_array(
         extension_metadata["ARROW:extension:metadata"] = json.dumps({"crs": crs_str})
 
     if geom_type == GeometryType.POINT:
-        parr = pa.FixedSizeListArray.from_arrays(coords.flatten(), len(dims))
+        parr = pa.FixedSizeListArray.from_arrays(coords.ravel("C"), len(dims))
         extension_metadata["ARROW:extension:name"] = "geoarrow.point"
         field = pa.field(
             field_name,
@@ -318,7 +318,7 @@ def construct_geometry_array(
     elif geom_type == GeometryType.LINESTRING:
         assert len(offsets) == 1, "Expected one offsets array"
         (geom_offsets,) = offsets
-        _parr = pa.FixedSizeListArray.from_arrays(coords.flatten(), len(dims))
+        _parr = pa.FixedSizeListArray.from_arrays(coords.ravel("C"), len(dims))
         parr = pa.ListArray.from_arrays(pa.array(geom_offsets), _parr)
         extension_metadata["ARROW:extension:name"] = "geoarrow.linestring"
         field = pa.field(
@@ -332,7 +332,7 @@ def construct_geometry_array(
     elif geom_type == GeometryType.POLYGON:
         assert len(offsets) == 2, "Expected two offsets arrays"
         ring_offsets, geom_offsets = offsets
-        _parr = pa.FixedSizeListArray.from_arrays(coords.flatten(), len(dims))
+        _parr = pa.FixedSizeListArray.from_arrays(coords.ravel("C"), len(dims))
         _parr1 = pa.ListArray.from_arrays(pa.array(ring_offsets), _parr)
         parr = pa.ListArray.from_arrays(pa.array(geom_offsets), _parr1)
         extension_metadata["ARROW:extension:name"] = "geoarrow.polygon"
@@ -347,7 +347,7 @@ def construct_geometry_array(
     elif geom_type == GeometryType.MULTIPOINT:
         assert len(offsets) == 1, "Expected one offsets array"
         (geom_offsets,) = offsets
-        _parr = pa.FixedSizeListArray.from_arrays(coords.flatten(), len(dims))
+        _parr = pa.FixedSizeListArray.from_arrays(coords.ravel("C"), len(dims))
         parr = pa.ListArray.from_arrays(pa.array(geom_offsets), _parr)
         extension_metadata["ARROW:extension:name"] = "geoarrow.multipoint"
         field = pa.field(
@@ -361,7 +361,7 @@ def construct_geometry_array(
     elif geom_type == GeometryType.MULTILINESTRING:
         assert len(offsets) == 2, "Expected two offsets arrays"
         ring_offsets, geom_offsets = offsets
-        _parr = pa.FixedSizeListArray.from_arrays(coords.flatten(), len(dims))
+        _parr = pa.FixedSizeListArray.from_arrays(coords.ravel("C"), len(dims))
         _parr1 = pa.ListArray.from_arrays(pa.array(ring_offsets), _parr)
         parr = pa.ListArray.from_arrays(pa.array(geom_offsets), _parr1)
         extension_metadata["ARROW:extension:name"] = "geoarrow.multilinestring"
@@ -376,7 +376,7 @@ def construct_geometry_array(
     elif geom_type == GeometryType.MULTIPOLYGON:
         assert len(offsets) == 3, "Expected three offsets arrays"
         ring_offsets, polygon_offsets, geom_offsets = offsets
-        _parr = pa.FixedSizeListArray.from_arrays(coords.flatten(), len(dims))
+        _parr = pa.FixedSizeListArray.from_arrays(coords.ravel("C"), len(dims))
         _parr1 = pa.ListArray.from_arrays(pa.array(ring_offsets), _parr)
         _parr2 = pa.ListArray.from_arrays(pa.array(polygon_offsets), _parr1)
         parr = pa.ListArray.from_arrays(pa.array(geom_offsets), _parr2)

lonboard/_geoarrow/ops/coord_layout.py

@@ -75,14 +75,14 @@ def _transpose_coords(arr: Union[pa.FixedSizeListArray, pa.StructArray]):
     if arr.type.num_fields == 2:
         x = arr.field("x").to_numpy()
         y = arr.field("y").to_numpy()
-        coords = np.column_stack([x, y]).flatten("C")
+        coords = np.column_stack([x, y]).ravel("C")
         return pa.FixedSizeListArray.from_arrays(coords, 2)
 
     if arr.type.num_fields == 3:
         x = arr.field("x").to_numpy()
         y = arr.field("y").to_numpy()
         z = arr.field("z").to_numpy()
-        coords = np.column_stack([x, y, z]).flatten("C")
+        coords = np.column_stack([x, y, z]).ravel("C")
         return pa.FixedSizeListArray.from_arrays(coords, 3)
 
     raise ValueError(f"Expected struct with 2 or 3 fields, got {arr.type.num_fields}")

lonboard/_geoarrow/ops/reproject.py

@@ -158,9 +158,7 @@ def _reproject_coords(arr: pa.FixedSizeListArray, transformer: Transformer):
         raise ValueError(f"Unexpected list size {list_size}")
 
     coord_field = pa.list_(pa.field(dims, pa.float64()), len(dims))
-    return pa.FixedSizeListArray.from_arrays(
-        output_np_arr.flatten("C"), type=coord_field
-    )
+    return pa.FixedSizeListArray.from_arrays(output_np_arr.ravel("C"), type=coord_field)
 
 
 def _reproject_chunk_nest_0(arr: pa.ListArray, transformer: Transformer):

lonboard/traits.py

@@ -272,7 +272,7 @@ def validate(
                     info="Color array must have 3 or 4 as its second dimension.",
                 )
 
-            return pa.FixedSizeListArray.from_arrays(value.flatten("C"), list_size)
+            return pa.FixedSizeListArray.from_arrays(value.ravel("C"), list_size)
 
         # Check for Arrow PyCapsule Interface
         # https://arrow.apache.org/docs/format/CDataInterface/PyCapsuleInterface.html
@@ -502,7 +502,7 @@ def validate(
                     info="Point array to have 2 or 3 as its second dimension",
                 )
 
-            return pa.FixedSizeListArray.from_arrays(value.flatten("C"), list_size)
+            return pa.FixedSizeListArray.from_arrays(value.ravel("C"), list_size)
 
         if isinstance(value, (pa.ChunkedArray, pa.Array)):
             if not pa.types.is_fixed_size_list(value.type):
@@ -681,7 +681,7 @@ def validate(self, obj, value) -> Union[float, pa.ChunkedArray, pa.DoubleArray]:
                     ),
                 )
 
-            return pa.FixedSizeListArray.from_arrays(value.flatten("C"), filter_size)
+            return pa.FixedSizeListArray.from_arrays(value.ravel("C"), filter_size)
 
         # Check for Arrow PyCapsule Interface
         # https://arrow.apache.org/docs/format/CDataInterface/PyCapsuleInterface.html
@@ -800,7 +800,7 @@ def validate(
                 )
                 value = value.astype(np.float32)
 
-            return pa.FixedSizeListArray.from_arrays(value.flatten("C"), 3)
+            return pa.FixedSizeListArray.from_arrays(value.ravel("C"), 3)
 
         # Check for Arrow PyCapsule Interface
         # https://arrow.apache.org/docs/format/CDataInterface/PyCapsuleInterface.html
@@ -932,7 +932,7 @@ def validate(
             if np.issubdtype(value.dtype, np.float64):
                 value = value.astype(np.float32)
 
-            return pa.FixedSizeListArray.from_arrays(value.flatten("C"), list_size)
+            return pa.FixedSizeListArray.from_arrays(value.ravel("C"), list_size)
 
         # Check for Arrow PyCapsule Interface
         # https://arrow.apache.org/docs/format/CDataInterface/PyCapsuleInterface.html