wip

pandas/core/arrays/sparse/array.py

@@ -892,13 +892,21 @@ def __getitem__(
  elif isinstance(key, tuple):
  data_slice = self.to_dense()[key]
  elif isinstance(key, slice):
- # special case to preserve dtypes
- if key == slice(None):
- return self.copy()
- # TODO: this logic is surely elsewhere
- # TODO: this could be more efficient
- indices = np.arange(len(self), dtype=np.int32)[key]
- return self.take(indices)
+ start = 0 if key.start is None else key.start
+ if start < 0:
+ start += len(self)
+
+ end = len(self) if key.step is None else key.stop
+ if end < 0:
+ end += len(self)
+
+ indices = self.sp_index.to_int_index().indices
+
+ keep_inds = np.flatnonzero((indices >= start) & (indices <= end))
+ sp_vals = self.sp_values[keep_inds]
+ sp_index = indices[keep_inds]
+ new_sp_index = make_sparse_index(len(self), sp_index, self.kind)
+ return type(self)._simple_new(sp_vals, new_sp_index, self.dtype)
  else:
  # TODO: I think we can avoid densifying when masking a
  # boolean SparseArray with another. Need to look at the
@@ -1762,10 +1770,10 @@ def make_sparse_index(length: int, indices, kind: Literal["integer"]) -> IntInde
 
 def make_sparse_index(length: int, indices, kind: SparseIndexKind) -> SparseIndex:
  index: SparseIndex
- if kind == "block" or isinstance(kind, BlockIndex):
+ if kind == "block":
  locs, lens = splib.get_blocks(indices)
  index = BlockIndex(length, locs, lens)
- elif kind == "integer" or isinstance(kind, IntIndex):
+ elif kind == "integer":
  index = IntIndex(length, indices)
  else: # pragma: no cover
  raise ValueError("must be block or integer type")

pandas/core/internals/blocks.py

@@ -2082,6 +2082,9 @@ def to_native_types(
  """convert to our native types format"""
  values = ensure_wrapped_if_datetimelike(values)
 
+ if is_sparse(values):
+ values = values.to_dense()
+
  if isinstance(values, (DatetimeArray, TimedeltaArray)):
  result = values._format_native_types(na_rep=na_rep, **kwargs)
  result = result.astype(object, copy=False)