Merge branch 'branch-24.02' into rework-dependencies

rapidsai · Jan 11, 2024 · cfd3edf · cfd3edf
2 parents 1b21663 + d26ea6d
commit cfd3edf
Show file tree

Hide file tree

Showing 6 changed files with 34 additions and 11 deletions.
diff --git a/python/cudf/cudf/core/_internals/where.py b/python/cudf/cudf/core/_internals/where.py
@@ -103,7 +103,7 @@ def _check_and_cast_columns_with_other(
         other = cudf.Scalar(other)
 
     if is_mixed_with_object_dtype(other, source_col) or (
-        is_bool_dtype(source_col) and not is_bool_dtype(common_dtype)
+        is_bool_dtype(source_dtype) and not is_bool_dtype(common_dtype)
     ):
         raise TypeError(mixed_err)
 

diff --git a/python/cudf/cudf/core/column/column.py b/python/cudf/cudf/core/column/column.py
@@ -25,6 +25,7 @@
 import numpy as np
 import pandas as pd
 import pyarrow as pa
+import pyarrow.compute as pc
 from numba import cuda
 from typing_extensions import Self
 
@@ -1997,11 +1998,19 @@ def as_column(
         return col
 
     elif isinstance(arbitrary, (pa.Array, pa.ChunkedArray)):
-        if isinstance(arbitrary, pa.lib.HalfFloatArray):
+        if pa.types.is_float16(arbitrary.type):
             raise NotImplementedError(
                 "Type casting from `float16` to `float32` is not "
                 "yet supported in pyarrow, see: "
-                "https://issues.apache.org/jira/browse/ARROW-3802"
+                "https://github.com/apache/arrow/issues/20213"
+            )
+        elif (nan_as_null is None or nan_as_null) and pa.types.is_floating(
+            arbitrary.type
+        ):
+            arbitrary = pc.if_else(
+                pc.is_nan(arbitrary),
+                pa.nulls(len(arbitrary), type=arbitrary.type),
+                arbitrary,
             )
         col = ColumnBase.from_arrow(arbitrary)
 

diff --git a/python/cudf/cudf/core/column/numerical.py b/python/cudf/cudf/core/column/numerical.py
@@ -272,13 +272,13 @@ def _binaryop(self, other: ColumnBinaryOperand, op: str) -> ColumnBase:
             out_dtype = "bool"
 
         if op in {"__and__", "__or__", "__xor__"}:
-            if is_float_dtype(self.dtype) or is_float_dtype(other):
+            if is_float_dtype(self.dtype) or is_float_dtype(other.dtype):
                 raise TypeError(
                     f"Operation 'bitwise {op[2:-2]}' not supported between "
                     f"{self.dtype.type.__name__} and "
                     f"{other.dtype.type.__name__}"
                 )
-            if is_bool_dtype(self.dtype) or is_bool_dtype(other):
+            if is_bool_dtype(self.dtype) or is_bool_dtype(other.dtype):
                 out_dtype = "bool"
 
         if (

diff --git a/python/cudf/cudf/core/dataframe.py b/python/cudf/cudf/core/dataframe.py
@@ -324,7 +324,7 @@ def _getitem_tuple_arg(self, arg):
                     tmp_arg[1],
                 )
 
-                if is_bool_dtype(tmp_arg[0]):
+                if is_bool_dtype(tmp_arg[0].dtype):
                     df = columns_df._apply_boolean_mask(
                         BooleanMask(tmp_arg[0], len(columns_df))
                     )
@@ -6032,7 +6032,7 @@ def _reduce(
             numeric_cols = (
                 name
                 for name in self._data.names
-                if is_numeric_dtype(self._data[name])
+                if is_numeric_dtype(self._data[name].dtype)
             )
             source = self._get_columns_by_label(numeric_cols)
             if source.empty:
@@ -6078,7 +6078,7 @@ def _reduce(
                     numeric_cols = (
                         name
                         for name in self._data.names
-                        if is_numeric_dtype(self._data[name])
+                        if is_numeric_dtype(self._data[name].dtype)
                     )
                     source = self._get_columns_by_label(numeric_cols)
                     if source.empty:

diff --git a/python/cudf/cudf/testing/testing.py b/python/cudf/cudf/testing/testing.py
@@ -232,10 +232,10 @@ def assert_column_equal(
     elif not (
         (
             not dtype_can_compare_equal_to_other(left.dtype)
-            and is_numeric_dtype(right)
+            and is_numeric_dtype(right.dtype)
         )
         or (
-            is_numeric_dtype(left)
+            is_numeric_dtype(left.dtype)
             and not dtype_can_compare_equal_to_other(right.dtype)
         )
     ):
@@ -245,7 +245,11 @@ def assert_column_equal(
                 left.isnull().values == right.isnull().values
             )
 
-            if columns_equal and not check_exact and is_numeric_dtype(left):
+            if (
+                columns_equal
+                and not check_exact
+                and is_numeric_dtype(left.dtype)
+            ):
                 # non-null values must be the same
                 columns_equal = cp.allclose(
                     left.apply_boolean_mask(

diff --git a/python/cudf/cudf/tests/test_series.py b/python/cudf/cudf/tests/test_series.py
@@ -2572,6 +2572,16 @@ def test_series_arrow_list_types_roundtrip():
             cudf.from_pandas(pdf)
 
 
+@pytest.mark.parametrize("klass", [cudf.Index, cudf.Series])
+@pytest.mark.parametrize(
+    "data", [pa.array([float("nan")]), pa.chunked_array([[float("nan")]])]
+)
+def test_nan_as_null_from_arrow_objects(klass, data):
+    result = klass(data, nan_as_null=True)
+    expected = klass(pa.array([None], type=pa.float64()))
+    assert_eq(result, expected)
+
+
 @pytest.mark.parametrize("reso", ["M", "ps"])
 @pytest.mark.parametrize("typ", ["M", "m"])
 def test_series_invalid_reso_dtype(reso, typ):