Fix typo and 0-d numpy array handling in binary operation(#6887)

Fixes typo and 0-d numpy array handling. When numpy scalar is used on lhs while executing binary operation, `__eq__` from numpy returns a 0-d array rather than scalar.

closes #6778

Authors:
  - Ramakrishna Prabhu <[email protected]>
  - Ram (Ramakrishna Prabhu) <[email protected]>

Approvers:
  - Keith Kraus

URL: #6887

CHANGELOG.md

@@ -183,6 +183,7 @@
 - PR #6861 Fix compile error in type_dispatch_benchmark.cu
 - PR #6869 Avoid dependency resolution failure in latest version of pip by explicitly specifying versions for dask and distributed
 - PR #6806 Force install of local conda artifacts
+- PR #6887 Fix typo and `0-d` numpy array handling in binary operation
 
 # cuDF 0.16.0 (21 Oct 2020)
 

python/cudf/cudf/core/column/categorical.py

@@ -890,6 +890,9 @@ def binary_operator(self, op, rhs, reflect=False):
 
     def normalize_binop_value(self, other):
 
+        if isinstance(other, np.ndarray) and other.ndim == 0:
+            other = other.item()
+
         ary = cudf.utils.utils.scalar_broadcast_to(
             self._encode(other), size=len(self), dtype=self.codes.dtype
         )

python/cudf/cudf/core/column/datetime.py

@@ -129,7 +129,11 @@ def get_dt_field(self, field):
     def normalize_binop_value(self, other):
         if isinstance(other, cudf.Scalar):
             return other
-        elif isinstance(other, dt.datetime):
+
+        if isinstance(other, np.ndarray) and other.ndim == 0:
+            other = other.item()
+
+        if isinstance(other, dt.datetime):
             other = np.datetime64(other)
         elif isinstance(other, dt.timedelta):
             other = np.timedelta64(other)

python/cudf/cudf/core/column/numerical.py

@@ -110,6 +110,8 @@ def normalize_binop_value(self, other):
             # expensive device-host transfer just to
             # adjust the dtype
             other = other.value
+        elif isinstance(other, np.ndarray) and other.ndim == 0:
+            other = other.item()
         other_dtype = np.min_scalar_type(other)
         if other_dtype.kind in {"b", "i", "u", "f"}:
             if isinstance(other, cudf.Scalar):

python/cudf/cudf/core/column/string.py

@@ -5014,6 +5014,11 @@ def normalize_binop_value(self, other):
                 other, size=len(self), dtype="object"
             )
             return col
+        elif isinstance(other, np.ndarray) and other.ndim == 0:
+            col = utils.scalar_broadcast_to(
+                other.item(), size=len(self), dtype="object"
+            )
+            return col
         else:
             raise TypeError(f"cannot broadcast {type(other)}")
 

python/cudf/cudf/core/column/timedelta.py

@@ -220,7 +220,11 @@ def binary_operator(self, op, rhs, reflect=False):
     def normalize_binop_value(self, other):
         if isinstance(other, cudf.Scalar):
             return other
-        elif isinstance(other, dt.timedelta):
+
+        if isinstance(other, np.ndarray) and other.ndim == 0:
+            other = other.item()
+
+        if isinstance(other, dt.timedelta):
             other = np.timedelta64(other)
         elif isinstance(other, pd.Timestamp):
             other = other.to_datetime64()

python/cudf/cudf/core/dataframe.py

@@ -1371,7 +1371,9 @@ def fallback(col, fn):
                         l_opr = self[col]
                 result[col] = op(l_opr, r_opr)
 
-        elif isinstance(other, (numbers.Number, cudf.Scalar)):
+        elif isinstance(other, (numbers.Number, cudf.Scalar)) or (
+            isinstance(other, np.ndarray) and other.ndim == 0
+        ):
             for col in self._data:
                 result[col] = op(self[col], other)
         else:

python/cudf/cudf/tests/test_binops.py

@@ -1459,3 +1459,35 @@ def test_scalar_power_invalid(dtype_l, dtype_r):
 
     with pytest.raises(TypeError):
         lval_gpu ** rval_gpu
+
+
+@pytest.mark.parametrize("frame", [cudf.Series, cudf.Index, cudf.DataFrame])
+@pytest.mark.parametrize(
+    "dtype", ["int", "str", "datetime64[s]", "timedelta64[s]", "category"]
+)
+def test_binops_with_lhs_numpy_scalar(frame, dtype):
+    data = [1, 2, 3, 4, 5]
+
+    data = (
+        frame({"a": data}, dtype=dtype)
+        if isinstance(frame, cudf.DataFrame)
+        else frame(data, dtype=dtype)
+    )
+
+    if dtype == "datetime64[s]":
+        val = np.dtype(dtype).type(4, "s")
+    elif dtype == "timedelta64[s]":
+        val = np.dtype(dtype).type(4, "s")
+    elif dtype == "category":
+        val = np.int64(4)
+    else:
+        val = np.dtype(dtype).type(4)
+
+    expected = val == data.to_pandas()
+    got = val == data
+
+    # In case of index, expected would be a numpy array
+    if isinstance(data, cudf.Index):
+        expected = pd.Index(expected)
+
+    utils.assert_eq(expected, got)