feat(python): improve Series/DataFrame init from existing Series/Data…

…Frame

py-polars/polars/dataframe/frame.py

@@ -83,6 +83,7 @@
     _post_apply_columns,
     arrow_to_pydf,
     dict_to_pydf,
+    frame_to_pydf,
     iterable_to_pydf,
     numpy_to_idxs,
     numpy_to_pydf,
@@ -381,6 +382,7 @@ def __init__(
                 orient=orient,
                 infer_schema_length=infer_schema_length,
             )
+
         elif isinstance(data, pl.Series):
             self._df = series_to_pydf(
                 data, schema=schema, schema_overrides=schema_overrides
@@ -413,6 +415,11 @@ def __init__(
                 orient=orient,
                 infer_schema_length=infer_schema_length,
             )
+
+        elif isinstance(data, pl.DataFrame):
+            self._df = frame_to_pydf(
+                data, schema=schema, schema_overrides=schema_overrides
+            )
         else:
             raise TypeError(
                 f"DataFrame constructor called with unsupported type {type(data).__name__!r}"

py-polars/polars/series/series.py

@@ -264,9 +264,12 @@ def __init__(
             )
 
         # Handle case where values are passed as the first argument
+        original_name: str | None = None
         if name is None:
             name = ""
-        elif not isinstance(name, str):
+        elif isinstance(name, str):
+            original_name = name
+        else:
             if values is None:
                 values = name
                 name = ""
@@ -277,12 +280,14 @@ def __init__(
             self._s = sequence_to_pyseries(
                 name, [], dtype=dtype, dtype_if_empty=dtype_if_empty
             )
-        elif isinstance(values, Series):
-            self._s = series_to_pyseries(name, values)
 
         elif isinstance(values, range):
             self._s = range_to_series(name, values, dtype=dtype)._s
 
+        elif isinstance(values, Series):
+            name = values.name if original_name is None else name
+            self._s = series_to_pyseries(name, values, dtype=dtype, strict=strict)
+
         elif isinstance(values, Sequence):
             self._s = sequence_to_pyseries(
                 name,
@@ -292,6 +297,7 @@ def __init__(
                 dtype_if_empty=dtype_if_empty,
                 nan_to_null=nan_to_null,
             )
+
         elif _check_for_numpy(values) and isinstance(values, np.ndarray):
             self._s = numpy_to_pyseries(
                 name, values, strict=strict, nan_to_null=nan_to_null
@@ -328,6 +334,17 @@ def __init__(
                 dtype_if_empty=dtype_if_empty,
                 strict=strict,
             )
+
+        elif isinstance(values, pl.DataFrame):
+            to_struct = len(values.columns) > 1
+            name = (
+                values.columns[0] if (original_name is None and not to_struct) else name
+            )
+            s = values.to_struct(name) if to_struct else values.to_series().rename(name)
+            if dtype is not None and dtype != s.dtype:
+                s = s.cast(dtype)
+            self._s = s._s
+
         else:
             raise TypeError(
                 f"Series constructor called with unsupported type {type(values).__name__!r}"

py-polars/polars/utils/_construction.py

@@ -164,9 +164,17 @@ def nt_unpack(obj: Any) -> Any:
 ################################
 
 
-def series_to_pyseries(name: str, values: Series) -> PySeries:
+def series_to_pyseries(
+    name: str,
+    values: Series,
+    *,
+    dtype: PolarsDataType | None = None,
+    strict: bool = True,
+) -> PySeries:
     """Construct a new PySeries from a Polars Series."""
     py_s = values._s.clone()
+    if dtype is not None and dtype != py_s.dtype():
+        py_s = py_s.cast(dtype, strict=strict)
     py_s.rename(name)
     return py_s
 
@@ -1603,6 +1611,9 @@ def series_to_pydf(
     schema_overrides: SchemaDict | None = None,
 ) -> PyDataFrame:
     """Construct a PyDataFrame from a Polars Series."""
+    if schema is None and schema_overrides is None:
+        return PyDataFrame([data._s])
+
     data_series = [data._s]
     series_name = [s.name() for s in data_series]
     column_names, schema_overrides = _unpack_schema(
@@ -1617,6 +1628,29 @@ def series_to_pydf(
     return PyDataFrame(data_series)
 
 
+def frame_to_pydf(
+    data: DataFrame,
+    schema: SchemaDefinition | None = None,
+    schema_overrides: SchemaDict | None = None,
+) -> PyDataFrame:
+    """Construct a PyDataFrame from an existing Polars DataFrame."""
+    if schema is None and schema_overrides is None:
+        return data._df.clone()
+
+    data_series = {c.name: c._s for c in data}
+    column_names, schema_overrides = _unpack_schema(
+        schema or data.columns, schema_overrides=schema_overrides
+    )
+    if schema_overrides:
+        existing_schema = data.schema
+        for name, new_dtype in schema_overrides.items():
+            if new_dtype != existing_schema[name]:
+                data_series[name] = data_series[name].cast(new_dtype, strict=True)
+
+    series_cols = _handle_columns_arg(list(data_series.values()), columns=column_names)
+    return PyDataFrame(series_cols)
+
+
 def iterable_to_pydf(
     data: Iterable[Any],
     schema: SchemaDefinition | None = None,

py-polars/tests/unit/datatypes/test_struct.py

@@ -10,7 +10,7 @@
 
 import polars as pl
 import polars.selectors as cs
-from polars.testing import assert_frame_equal
+from polars.testing import assert_frame_equal, assert_series_equal
 
 if TYPE_CHECKING:
     from polars.datatypes import PolarsDataType
@@ -165,12 +165,16 @@ def test_struct_function_expansion() -> None:
         {"a": [1, 2, 3, 4], "b": ["one", "two", "three", "four"], "c": [9, 8, 7, 6]}
     )
     struct_schema = {"a": pl.UInt32, "b": pl.String}
-    s = df.with_columns(pl.struct(pl.col(["a", "b"]), schema=struct_schema))["a"]
+    dfs = df.with_columns(pl.struct(pl.col(["a", "b"]), schema=struct_schema))
+    s = dfs["a"]
 
     assert isinstance(s, pl.Series)
     assert s.struct.fields == ["a", "b"]
     assert pl.Struct(struct_schema) == s.to_frame().schema["a"]
 
+    assert_series_equal(s, pl.Series(dfs.select("a")))
+    assert_frame_equal(dfs, pl.DataFrame(dfs))
+
 
 def test_nested_struct() -> None:
     df = pl.DataFrame({"d": [1, 2, 3], "e": ["foo", "bar", "biz"]})
@@ -187,11 +191,11 @@ def test_nested_struct() -> None:
 
 
 def test_struct_to_pandas() -> None:
-    df = pd.DataFrame([{"a": {"b": {"c": 2}}}])
-    pl_df = pl.from_pandas(df)
+    pdf = pd.DataFrame([{"a": {"b": {"c": 2}}}])
+    df = pl.from_pandas(pdf)
 
-    assert isinstance(pl_df.dtypes[0], pl.datatypes.Struct)
-    assert pl_df.to_pandas().equals(df)
+    assert isinstance(df.dtypes[0], pl.datatypes.Struct)
+    assert df.to_pandas().equals(pdf)
 
 
 def test_struct_logical_types_to_pandas() -> None:

py-polars/tests/unit/test_constructors.py

@@ -711,6 +711,28 @@ def test_init_arrow() -> None:
         pl.DataFrame(pa.table({"a": [1, 2, 3], "b": [4, 5, 6]}), schema=["c", "d", "e"])
 
 
+def test_init_from_frame() -> None:
+    df1 = pl.DataFrame({"id": [0, 1], "misc": ["a", "b"], "val": [-10, 10]})
+    assert_frame_equal(df1, pl.DataFrame(df1))
+
+    df2 = pl.DataFrame(df1, schema=["a", "b", "c"])
+    assert_frame_equal(df2, pl.DataFrame(df2))
+
+    df3 = pl.DataFrame(df1, schema=["a", "b", "c"], schema_overrides={"val": pl.Int8})
+    assert_frame_equal(df3, pl.DataFrame(df3))
+
+    assert df1.schema == {"id": pl.Int64, "misc": pl.String, "val": pl.Int64}
+    assert df2.schema == {"a": pl.Int64, "b": pl.String, "c": pl.Int64}
+    assert df3.schema == {"a": pl.Int64, "b": pl.String, "c": pl.Int8}
+    assert df1.rows() == df2.rows() == df3.rows()
+
+    s1 = pl.Series("s", df3)
+    s2 = pl.Series(df3)
+
+    assert s1.name == "s"
+    assert s2.name == ""
+
+
 def test_init_series() -> None:
     # List of Series
     df = pl.DataFrame([pl.Series("a", [1, 2, 3]), pl.Series("b", [4, 5, 6])])
@@ -730,9 +752,9 @@ def test_init_series() -> None:
 
     # List of unnamed Series
     df = pl.DataFrame([pl.Series([1, 2, 3]), pl.Series([4, 5, 6])])
-    expected = pl.DataFrame(
-        [pl.Series("column_0", [1, 2, 3]), pl.Series("column_1", [4, 5, 6])]
-    )
+    col0 = pl.Series("column_0", [1, 2, 3])
+    col1 = pl.Series("column_1", [4, 5, 6])
+    expected = pl.DataFrame([col0, col1])
     assert_frame_equal(df, expected)
 
     df = pl.DataFrame([pl.Series([0.0]), pl.Series([1.0])])
@@ -763,8 +785,16 @@ def test_init_series() -> None:
     s2 = pl.Series([[[2, 2]]], dtype=pl.List(pl.List(pl.UInt8)))
     assert s2.dtype == pl.List(pl.List(pl.UInt8))
 
-    s3 = pl.Series(dtype=pl.List(pl.List(pl.UInt8)))
-    assert s3.dtype == pl.List(pl.List(pl.UInt8))
+    nested_dtype = pl.List(pl.List(pl.UInt8))
+    s3 = pl.Series("x", dtype=nested_dtype)
+    s4 = pl.Series(s3)
+    for s in (s3, s4):
+        assert s.dtype == nested_dtype
+        assert s.to_list() == []
+        assert s.name == "x"
+
+    s5 = pl.Series("", df, dtype=pl.Int8)
+    assert_series_equal(s5, pl.Series("", [1, 2, 3], dtype=pl.Int8))
 
 
 def test_init_seq_of_seq() -> None: