Handle some corner-cases in indexing with boolean masks

These must be treated specially and not accidentally converted to
integers before indexing.

- Closes #13015
- Closes #13265
- Closes #13270

python/cudf/cudf/core/column_accessor.py

@@ -19,6 +19,7 @@
 
 import pandas as pd
 from packaging.version import Version
+from pandas.api.types import is_bool
 
 import cudf
 from cudf.core import column
@@ -359,7 +360,8 @@ def get_labels_by_index(self, index: Any) -> tuple:
 
         Parameters
         ----------
-        index : integer, integer slice, or list-like of integers
+        index : integer, integer slice, boolean mask,
+            or list-like of integers
             The column indexes.
 
         Returns
@@ -371,6 +373,14 @@ def get_labels_by_index(self, index: Any) -> tuple:
             return self.names[start:stop:step]
         elif pd.api.types.is_integer(index):
             return (self.names[index],)
+        elif index and all(map(is_bool, index)):
+            if len(index) != len(self.names):
+                raise IndexError("Invalid boolean mask for column selection")
+            if isinstance(index, (pd.Series, cudf.Series)):
+                # Don't allow iloc indexing with series
+                raise IndexError("Cannot use Series object for iloc indexing")
+            # TODO: Doesn't handle on-device columns
+            return tuple(n for n, keep in zip(self.names, index) if keep)
         else:
             return tuple(self.names[i] for i in index)
 
@@ -381,7 +391,8 @@ def select_by_index(self, index: Any) -> ColumnAccessor:
 
         Parameters
         ----------
-        key : integer, integer slice, or list-like of integers
+        key : integer, integer slice, boolean mask,
+            or list-like of integers
 
         Returns
         -------
@@ -462,7 +473,17 @@ def set_by_label(self, key: Any, value: Any, validate: bool = True):
         self._clear_cache()
 
     def _select_by_label_list_like(self, key: Any) -> ColumnAccessor:
-        data = {k: self._grouped_data[k] for k in key}
+        # Might be a generator
+        key = tuple(key)
+        # Special-casing for boolean mask
+        if key and all(map(is_bool, key)):
+            data = dict(
+                item
+                for item, keep in zip(self._grouped_data.items(), key)
+                if keep
+            )
+        else:
+            data = {k: self._grouped_data[k] for k in key}
         if self.multiindex:
             data = _to_flat_dict(data)
         return self.__class__(

python/cudf/cudf/core/indexed_frame.py

@@ -2904,7 +2904,8 @@ def _apply_boolean_mask(self, boolean_mask):
         boolean_mask = cudf.core.column.as_column(boolean_mask)
         if not is_bool_dtype(boolean_mask.dtype):
             raise ValueError("boolean_mask is not boolean type.")
-
+        if (bn := len(boolean_mask)) != (n := len(self)):
+            raise IndexError(f"Boolean index has wrong length: {bn} not {n}")
         return self._from_columns_like_self(
             libcudf.stream_compaction.apply_boolean_mask(
                 list(self._index._columns + self._columns), boolean_mask

python/cudf/cudf/core/single_column_frame.py

@@ -403,6 +403,10 @@ def _get_elements_from_column(self, arg) -> Union[ScalarLike, ColumnBase]:
             if is_integer_dtype(arg.dtype):
                 return self._column.take(arg)
             if is_bool_dtype(arg.dtype):
+                if (bn := len(arg)) != (n := len(self)):
+                    raise IndexError(
+                        f"Boolean index has wrong length: {bn} not {n}"
+                    )
                 return self._column.apply_boolean_mask(arg)
             raise NotImplementedError(f"Unknown indexer {type(arg)}")
 

python/cudf/cudf/tests/test_indexing.py

@@ -579,7 +579,6 @@ def test_dataframe_series_loc_multiindex(obj):
 
 @pytest.mark.parametrize("nelem", [2, 5, 20, 100])
 def test_series_iloc(nelem):
-
     # create random cudf.Series
     np.random.seed(12)
     ps = pd.Series(np.random.sample(nelem))
@@ -1265,7 +1264,6 @@ def test_iloc_categorical_index(index):
 )
 @pytest.mark.parametrize("is_dataframe", [True, False])
 def test_loc_datetime_index(sli, is_dataframe):
-
     if is_dataframe is True:
         pd_data = pd.DataFrame(
             {"a": [1, 2, 3]},
@@ -1685,3 +1683,33 @@ def test_loc_single_row_from_slice():
     pdf = pd.DataFrame({"a": [10, 20, 30], "b": [1, 2, 3]}).set_index("a")
     df = cudf.from_pandas(pdf)
     assert_eq(pdf.loc[5:10], df.loc[5:10])
+
+
+@pytest.mark.parametrize("indexer", ["loc", "iloc"])
+@pytest.mark.parametrize(
+    "mask",
+    [[False, True], [False, False, True, True, True]],
+    ids=["too-short", "too-long"],
+)
+def test_boolean_mask_wrong_length(indexer, mask):
+    s = pd.Series([1, 2, 3, 4])
+
+    indexee = getattr(s, indexer)
+    with pytest.raises(IndexError):
+        indexee[mask]
+
+    c = cudf.from_pandas(s)
+    indexee = getattr(c, indexer)
+    with pytest.raises(IndexError):
+        indexee[mask]
+
+
+@pytest.mark.parametrize("indexer", ["loc", "iloc"])
+def test_boolean_mask_columns(indexer):
+    df = pd.DataFrame(np.zeros((3, 3)))
+    cdf = cudf.from_pandas(df)
+
+    expect = getattr(df, indexer)[:, [True, False, True]]
+    got = getattr(cdf, indexer)[:, [True, False, True]]
+
+    assert_eq(expect, got)