Interval index and interval_range (#7182)

This PR is a follow up to PR #6984. It adds the IntervalIndex and interval_range classes to CuDF. IntervalIndex is needed for the `cut` and `histogram` methods and closes issue #5376.   

The main files to take note of in this PR are `index.py` and the two cudf `__init__.py` files. The rest of the file changes are related to PR6984 for creating the interval dtype.

Authors:
  - Marlene  (https://github.com/marlenezw)
  - Michael Wang (https://github.com/isVoid)

Approvers:
  - Keith Kraus (https://github.com/kkraus14)

URL: #7182

python/cudf/cudf/__init__.py

@@ -18,6 +18,8 @@
 from cudf.core import (
     NA,
     CategoricalIndex,
+    interval_range,
+    IntervalIndex,
     DataFrame,
     DatetimeIndex,
     Float32Index,

python/cudf/cudf/core/__init__.py

@@ -5,6 +5,8 @@
 from cudf.core.dataframe import DataFrame, from_pandas, merge
 from cudf.core.index import (
     CategoricalIndex,
+    interval_range,
+    IntervalIndex,
     DatetimeIndex,
     Float32Index,
     Float64Index,

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

@@ -42,6 +42,7 @@
 from cudf.core.abc import Serializable
 from cudf.core.buffer import Buffer
 from cudf.core.dtypes import CategoricalDtype
+from cudf.core.dtypes import IntervalDtype
 from cudf.utils import ioutils, utils
 from cudf.utils.dtypes import (
     NUMERIC_TYPES,
@@ -1046,11 +1047,7 @@ def astype(self, dtype: Dtype, **kwargs) -> ColumnBase:
                 )
             return self
         elif is_interval_dtype(self.dtype):
-            if not self.dtype == dtype:
-                raise NotImplementedError(
-                    "Casting interval columns not currently supported"
-                )
-            return self
+            return self.as_interval_column(dtype, **kwargs)
         elif is_decimal_dtype(dtype):
             return self.as_decimal_column(dtype, **kwargs)
         elif np.issubdtype(dtype, np.datetime64):
@@ -1113,6 +1110,11 @@ def as_datetime_column(
     ) -> "cudf.core.column.DatetimeColumn":
         raise NotImplementedError
 
+    def as_interval_column(
+        self, dtype: Dtype, **kwargs
+    ) -> "cudf.core.column.IntervalColumn":
+        raise NotImplementedError
+
     def as_timedelta_column(
         self, dtype: Dtype, **kwargs
     ) -> "cudf.core.column.TimeDeltaColumn":
@@ -1633,6 +1635,15 @@ def build_column(
             null_count=null_count,
             children=children,
         )
+    elif is_interval_dtype(dtype):
+        return cudf.core.column.IntervalColumn(
+            dtype=dtype,
+            mask=mask,
+            size=size,
+            offset=offset,
+            children=children,
+            null_count=null_count,
+        )
     elif is_struct_dtype(dtype):
         if size is None:
             raise TypeError("Must specify size")
@@ -1714,6 +1725,52 @@ def build_categorical_column(
     return cast("cudf.core.column.CategoricalColumn", result)
 
 
+def build_interval_column(
+    left_col,
+    right_col,
+    mask=None,
+    size=None,
+    offset=0,
+    null_count=None,
+    closed="right",
+):
+    """
+    Build an IntervalColumn
+
+    Parameters
+    ----------
+    left_col : Column
+        Column of values representing the left of the interval
+    right_col : Column
+        Column of representing the right of the interval
+    mask : Buffer
+        Null mask
+    size : int, optional
+    offset : int, optional
+    closed : {"left", "right", "both", "neither"}, default "right"
+            Whether the intervals are closed on the left-side, right-side,
+            both or neither.
+    """
+    left = as_column(left_col)
+    right = as_column(right_col)
+    if closed not in {"left", "right", "both", "neither"}:
+        closed = "right"
+    if type(left_col) is not list:
+        dtype = IntervalDtype(left_col.dtype, closed)
+    else:
+        dtype = IntervalDtype("int64", closed)
+    size = len(left)
+    return build_column(
+        data=None,
+        dtype=dtype,
+        mask=mask,
+        size=size,
+        offset=offset,
+        null_count=null_count,
+        children=(left, right),
+    )
+
+
 def as_column(
     arbitrary: Any,
     nan_as_null: bool = None,
@@ -2102,7 +2159,7 @@ def as_column(
                     data = as_column(sr, nan_as_null=nan_as_null)
                 elif is_interval_dtype(dtype):
                     sr = pd.Series(arbitrary, dtype="interval")
-                    data = as_column(sr, nan_as_null=nan_as_null)
+                    data = as_column(sr, nan_as_null=nan_as_null, dtype=dtype)
                 else:
                     data = as_column(
                         _construct_array(arbitrary, dtype),

python/cudf/cudf/core/column/interval.py

@@ -2,6 +2,8 @@
 import pyarrow as pa
 import cudf
 from cudf.core.column import StructColumn
+from cudf.core.dtypes import IntervalDtype
+from cudf.utils.dtypes import is_interval_dtype
 
 
 class IntervalColumn(StructColumn):
@@ -38,7 +40,7 @@ def closed(self):
     def from_arrow(self, data):
         new_col = super().from_arrow(data.storage)
         size = len(data)
-        dtype = cudf.core.dtypes.IntervalDtype.from_arrow(data.type)
+        dtype = IntervalDtype.from_arrow(data.type)
         mask = data.buffers()[0]
         if mask is not None:
             mask = cudf.utils.utils.pa_mask_buffer_to_mask(mask, len(data))
@@ -60,14 +62,17 @@ def from_arrow(self, data):
 
     def to_arrow(self):
         typ = self.dtype.to_arrow()
-        return pa.ExtensionArray.from_storage(typ, super().to_arrow())
+        struct_arrow = super().to_arrow()
+        if len(struct_arrow) == 0:
+            # struct arrow is pa.struct array with null children types
+            # we need to make sure its children have non-null type
+            struct_arrow = pa.array([], typ.storage_type)
+        return pa.ExtensionArray.from_storage(typ, struct_arrow)
 
     def from_struct_column(self, closed="right"):
         return IntervalColumn(
             size=self.size,
-            dtype=cudf.core.dtypes.IntervalDtype(
-                self.dtype.fields["left"], closed
-            ),
+            dtype=IntervalDtype(self.dtype.fields["left"], closed),
             mask=self.base_mask,
             offset=self.offset,
             null_count=self.null_count,
@@ -80,12 +85,28 @@ def copy(self, deep=True):
         struct_copy = super().copy(deep=deep)
         return IntervalColumn(
             size=struct_copy.size,
-            dtype=cudf.core.dtypes.IntervalDtype(
-                struct_copy.dtype.fields["left"], closed
-            ),
+            dtype=IntervalDtype(struct_copy.dtype.fields["left"], closed),
             mask=struct_copy.base_mask,
             offset=struct_copy.offset,
             null_count=struct_copy.null_count,
             children=struct_copy.base_children,
             closed=closed,
         )
+
+    def as_interval_column(self, dtype, **kwargs):
+        if is_interval_dtype(dtype):
+            # a user can directly input the string `interval` as the dtype
+            # when creating an interval series or interval dataframe
+            if dtype == "interval":
+                dtype = IntervalDtype(self.dtype.fields["left"], self.closed)
+            return IntervalColumn(
+                size=self.size,
+                dtype=dtype,
+                mask=self.mask,
+                offset=self.offset,
+                null_count=self.null_count,
+                children=self.children,
+                closed=dtype.closed,
+            )
+        else:
+            raise ValueError("dtype must be IntervalDtype")