Disallow cudf.Series to accept column in favor of ._from_column (#1…

…6454)

`cudf.Series` is a public constructor that happens to accept a private `ColumnBase` object. Many ops return Columns and is natural to want to reconstruct a `Series`.

This PR adds a `SingleColumnFrame._from_column` classmethod for instances where we need to wrap a new column in an `Index` or `Series`. This constructor also passes some unneeded validation in `ColumnAccessor` and `Series`

Authors:
  - Matthew Roeschke (https://github.com/mroeschke)
  - GALI PREM SAGAR (https://github.com/galipremsagar)

Approvers:
  - GALI PREM SAGAR (https://github.com/galipremsagar)

URL: #16454

python/cudf/cudf/core/byte_pair_encoding.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2023, NVIDIA CORPORATION.
+# Copyright (c) 2023-2024, NVIDIA CORPORATION.
 
 from __future__ import annotations
 
@@ -27,7 +27,7 @@ class BytePairEncoder:
     def __init__(self, merges_pair: "cudf.Series"):
         self.merge_pairs = cpp_merge_pairs(merges_pair._column)
 
-    def __call__(self, text, separator: str = " "):
+    def __call__(self, text, separator: str = " ") -> cudf.Series:
         """
 
         Parameters
@@ -56,4 +56,4 @@ def __call__(self, text, separator: str = " "):
         sep = cudf.Scalar(separator, dtype="str")
         result = cpp_byte_pair_encoding(text._column, self.merge_pairs, sep)
 
-        return cudf.Series(result)
+        return cudf.Series._from_column(result)

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

@@ -123,7 +123,7 @@ def categories(self) -> "cudf.core.index.Index":
         return self._column.dtype.categories
 
     @property
-    def codes(self) -> "cudf.Series":
+    def codes(self) -> cudf.Series:
         """
         Return Series of codes as well as the index.
         """
@@ -132,7 +132,7 @@ def codes(self) -> "cudf.Series":
             if isinstance(self._parent, cudf.Series)
             else None
         )
-        return cudf.Series(self._column.codes, index=index)
+        return cudf.Series._from_column(self._column.codes, index=index)
 
     @property
     def ordered(self) -> bool:
@@ -918,7 +918,7 @@ def find_and_replace(
             )
             cur_categories = replaced.categories
             new_categories = cur_categories.apply_boolean_mask(
-                ~cudf.Series(cur_categories.isin(drop_values))
+                cur_categories.isin(drop_values).unary_operator("not")
             )
             replaced = replaced._set_categories(new_categories)
             df = df.dropna(subset=["new"])
@@ -943,7 +943,7 @@ def find_and_replace(
         # If a category is being replaced by an existing one, we
         # want to map it to None. If it's totally new, we want to
         # map it to the new label it is to be replaced by
-        dtype_replace = cudf.Series._from_data({None: replacement_col})
+        dtype_replace = cudf.Series._from_column(replacement_col)
         dtype_replace[dtype_replace.isin(cats_col)] = None
         new_cats_col = cats_col.find_and_replace(
             to_replace_col, dtype_replace._column
@@ -1273,12 +1273,8 @@ def _categories_equal(
             return False
         # if order doesn't matter, sort before the equals call below
         if not ordered:
-            cur_categories = cudf.Series(cur_categories).sort_values(
-                ignore_index=True
-            )
-            new_categories = cudf.Series(new_categories).sort_values(
-                ignore_index=True
-            )
+            cur_categories = cur_categories.sort_values()
+            new_categories = new_categories.sort_values()
         return cur_categories.equals(new_categories)
 
     def _set_categories(

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

@@ -7,6 +7,8 @@
 from typing_extensions import Literal
 
 import cudf
+import cudf.core.column
+import cudf.core.column_accessor
 from cudf.utils.utils import NotIterable
 
 ParentType = Union["cudf.Series", "cudf.core.index.Index"]
@@ -84,14 +86,11 @@ def _return_or_inplace(
                         data=table, index=self._parent.index
                     )
             elif isinstance(self._parent, cudf.Series):
-                if retain_index:
-                    return cudf.Series(
-                        new_col,
-                        name=self._parent.name,
-                        index=self._parent.index,
-                    )
-                else:
-                    return cudf.Series(new_col, name=self._parent.name)
+                return cudf.Series._from_column(
+                    new_col,
+                    name=self._parent.name,
+                    index=self._parent.index if retain_index else None,
+                )
             elif isinstance(self._parent, cudf.BaseIndex):
                 return cudf.Index(new_col, name=self._parent.name)
             else:

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

@@ -555,11 +555,8 @@ def can_cast_safely(self, to_dtype: DtypeObj) -> bool:
 
                 if self.dtype.kind == "f":
                     # Exclude 'np.inf', '-np.inf'
-                    s = cudf.Series(self)
-                    # TODO: replace np.inf with cudf scalar when
-                    # https://github.com/rapidsai/cudf/pull/6297 merges
-                    non_infs = s[~((s == np.inf) | (s == -np.inf))]
-                    col = non_infs._column
+                    not_inf = (self != np.inf) & (self != -np.inf)
+                    col = self.apply_boolean_mask(not_inf)
                 else:
                     col = self
 
@@ -599,8 +596,7 @@ def can_cast_safely(self, to_dtype: DtypeObj) -> bool:
             else:
                 filled = self.fillna(0)
                 return (
-                    cudf.Series(filled).astype(to_dtype).astype(filled.dtype)
-                    == cudf.Series(filled)
+                    filled.astype(to_dtype).astype(filled.dtype) == filled
                 ).all()
 
         # want to cast float to int:
@@ -615,7 +611,7 @@ def can_cast_safely(self, to_dtype: DtypeObj) -> bool:
             # NOTE(seberg): it would make sense to limit to the mantissa range.
             if (float(self.min()) >= min_) and (float(self.max()) <= max_):
                 filled = self.fillna(0)
-                return (cudf.Series(filled) % 1 == 0).all()
+                return (filled % 1 == 0).all()
             else:
                 return False
 

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

@@ -358,7 +358,7 @@ def cat(self, others=None, sep=None, na_rep=None):
             )
 
         if len(data) == 1 and data.null_count == 1:
-            data = [""]
+            data = cudf.core.column.as_column("", length=len(data))
         # We only want to keep the index if we are adding something to each
         # row, not if we are joining all the rows into a single string.
         out = self._return_or_inplace(data, retain_index=others is not None)
@@ -3623,7 +3623,7 @@ def findall(self, pat: str, flags: int = 0) -> SeriesOrIndex:
         data = libstrings.findall(self._column, pat, flags)
         return self._return_or_inplace(data)
 
-    def find_multiple(self, patterns: SeriesOrIndex) -> "cudf.Series":
+    def find_multiple(self, patterns: SeriesOrIndex) -> cudf.Series:
         """
         Find all first occurrences of patterns in the Series/Index.
 
@@ -3679,12 +3679,12 @@ def find_multiple(self, patterns: SeriesOrIndex) -> "cudf.Series":
                 f"got: {patterns_column.dtype}"
             )
 
-        return cudf.Series(
+        return cudf.Series._from_column(
             libstrings.find_multiple(self._column, patterns_column),
+            name=self._parent.name,
             index=self._parent.index
             if isinstance(self._parent, cudf.Series)
             else self._parent,
-            name=self._parent.name,
         )
 
     def isempty(self) -> SeriesOrIndex:
@@ -4376,14 +4376,9 @@ def code_points(self) -> SeriesOrIndex:
         2    99
         dtype: int32
         """
-
-        new_col = libstrings.code_points(self._column)
-        if isinstance(self._parent, cudf.Series):
-            return cudf.Series(new_col, name=self._parent.name)
-        elif isinstance(self._parent, cudf.BaseIndex):
-            return cudf.Index(new_col, name=self._parent.name)
-        else:
-            return new_col
+        return self._return_or_inplace(
+            libstrings.code_points(self._column), retain_index=False
+        )
 
     def translate(self, table: dict) -> SeriesOrIndex:
         """
@@ -4694,7 +4689,9 @@ def character_tokenize(self) -> SeriesOrIndex:
         if isinstance(self._parent, cudf.Series):
             lengths = self.len().fillna(0)
             index = self._parent.index.repeat(lengths)
-            return cudf.Series(result_col, name=self._parent.name, index=index)
+            return cudf.Series._from_column(
+                result_col, name=self._parent.name, index=index
+            )
         elif isinstance(self._parent, cudf.BaseIndex):
             return cudf.Index(result_col, name=self._parent.name)
         else: