[REVIEW] Fix issue related to .loc incase of DatetimeIndex

CHANGELOG.md

@@ -16,6 +16,7 @@
 
 ## Bug Fixes
 
+- PR #6073 Fix issue related to `.loc` incase of `DatetimeIndex`
 - PR #6081 Fix issue where fsspec thinks it has a protocol string
 
 

conda/environments/cudf_dev_cuda10.0.yml

@@ -13,7 +13,7 @@ dependencies:
   - cmake>=3.14
   - cmake_setuptools>=0.1.3
   - python>=3.6,<3.8
-  - numba>=0.49.0
+  - numba>=0.49.0,!=0.51.0
   - numpy
   - pandas>=1.0,<1.2.0dev0
   - pyarrow=0.17.1
@@ -50,6 +50,7 @@ dependencies:
   - hypothesis
   - sphinx-markdown-tables
   - sphinx-copybutton
+  - packaging
   - pip:
       - git+https://github.com/dask/dask.git
       - git+https://github.com/dask/distributed.git

conda/environments/cudf_dev_cuda10.1.yml

@@ -13,7 +13,7 @@ dependencies:
   - cmake>=3.14
   - cmake_setuptools>=0.1.3
   - python>=3.6,<3.8
-  - numba>=0.49.0
+  - numba>=0.49.0,!=0.51.0
   - numpy
   - pandas>=1.0,<1.2.0dev0
   - pyarrow=0.17.1
@@ -50,6 +50,7 @@ dependencies:
   - hypothesis
   - sphinx-markdown-tables
   - sphinx-copybutton
+  - packaging
   - pip:
       - git+https://github.com/dask/dask.git
       - git+https://github.com/dask/distributed.git

conda/environments/cudf_dev_cuda10.2.yml

@@ -13,7 +13,7 @@ dependencies:
   - cmake>=3.14
   - cmake_setuptools>=0.1.3
   - python>=3.6,<3.8
-  - numba>=0.49.0
+  - numba>=0.49,!=0.51.0
   - numpy
   - pandas>=1.0,<1.2.0dev0
   - pyarrow=0.17.1
@@ -50,6 +50,7 @@ dependencies:
   - hypothesis
   - sphinx-markdown-tables
   - sphinx-copybutton
+  - packaging
   - pip:
       - git+https://github.com/dask/dask.git
       - git+https://github.com/dask/distributed.git

python/cudf/cudf/core/_compat.py

@@ -1,4 +1,5 @@
 # Copyright (c) 2020, NVIDIA CORPORATION.
+
 import pandas as pd
 from packaging import version
 

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

@@ -259,15 +259,15 @@ def find_first_value(self, value, closest=False):
         Returns offset of first value that matches
         """
         value = pd.to_datetime(value)
-        value = column.as_column(value).as_numerical[0]
+        value = column.as_column(value, dtype=self.dtype).as_numerical[0]
         return self.as_numerical.find_first_value(value, closest=closest)
 
     def find_last_value(self, value, closest=False):
         """
         Returns offset of last value that matches
         """
         value = pd.to_datetime(value)
-        value = column.as_column(value).as_numerical[0]
+        value = column.as_column(value, dtype=self.dtype).as_numerical[0]
         return self.as_numerical.find_last_value(value, closest=closest)
 
     @property

python/cudf/cudf/tests/test_indexing.py

@@ -362,6 +362,22 @@ def test_series_loc_numerical():
     assert_eq(ps.loc[[5, 8, 9]], gs.loc[cupy.array([5, 8, 9])])
 
 
+def test_series_loc_float_index():
+    ps = pd.Series([1, 2, 3, 4, 5], index=[5.43, 6.34, 7.34, 8.0, 9.1])
+    gs = Series.from_pandas(ps)
+
+    assert_eq(ps.loc[5.43], gs.loc[5.43])
+    assert_eq(ps.loc[8], gs.loc[8])
+    assert_eq(ps.loc[6.1:8], gs.loc[6.1:8])
+    assert_eq(ps.loc[:7.1], gs.loc[:7.1])
+    assert_eq(ps.loc[6.345:], gs.loc[6.345:])
+    assert_eq(ps.loc[::2], gs.loc[::2])
+    assert_eq(
+        ps.loc[[True, False, True, False, True]],
+        gs.loc[[True, False, True, False, True]],
+    )
+
+
 def test_series_loc_string():
     ps = pd.Series(
         [1, 2, 3, 4, 5], index=["one", "two", "three", "four", "five"]
@@ -411,11 +427,12 @@ def test_series_loc_datetime():
     assert_eq(ps.loc["2001-01-02":], gs.loc["2001-01-02":], **kwargs)
     assert_eq(ps.loc[:"2001-01-04"], gs.loc[:"2001-01-04"], **kwargs)
     assert_eq(ps.loc[::2], gs.loc[::2], **kwargs)
-    #
-    # assert_eq(ps.loc[['2001-01-01', '2001-01-04', '2001-01-05']],
-    #           gs.loc[['2001-01-01', '2001-01-04', '2001-01-05']])
-    # looks like a bug in Pandas doesn't let us check for the above,
-    # so instead:
+
+    assert_eq(
+        ps.loc[["2001-01-01", "2001-01-04", "2001-01-05"]],
+        gs.loc[["2001-01-01", "2001-01-04", "2001-01-05"]],
+    )
+
     assert_eq(
         ps.loc[
             [
@@ -438,6 +455,12 @@ def test_series_loc_datetime():
         **kwargs,
     )
 
+    just_less_than_max = ps.index.max() - pd.Timedelta("5m")
+
+    assert_eq(
+        ps.loc[:just_less_than_max], gs.loc[:just_less_than_max], **kwargs
+    )
+
 
 def test_series_loc_categorical():
     ps = pd.Series(

python/cudf/cudf/utils/cudautils.py

@@ -105,7 +105,12 @@ def gpu_mark_found_int(arr, val, out, not_found):
 def gpu_mark_found_float(arr, val, out, not_found):
     i = cuda.grid(1)
     if i < arr.size:
-        if check_equals_float(arr[i], val):
+        # TODO: Remove val typecast to float(val)
+        # once numba minimum version is pinned
+        # at 0.51.1, this will have a very slight
+        # performance improvement. Related
+        # discussion in : https://github.com/rapidsai/cudf/pull/6073
+        if check_equals_float(arr[i], float(val)):
             out[i] = i
         else:
             out[i] = not_found
@@ -144,7 +149,7 @@ def find_index_of_val(arr, val, mask=None, compare="eq"):
     mask : mask of the array
     compare: str ('gt', 'lt', or 'eq' (default))
     """
-    found = cuda.device_array_like(arr)
+    found = cuda.device_array(shape=(arr.shape), dtype="int32")
     if found.size > 0:
         if compare == "gt":
             gpu_mark_gt.forall(found.size)(arr, val, found, arr.size)