feat(python): handle Series init from python sequence of numpy arrays (…

…#5918)

py-polars/polars/datatypes_constructor.py

@@ -3,6 +3,7 @@
 from typing import Any, Callable, Sequence
 
 from polars.datatypes import (
+    DTYPE_TEMPORAL_UNITS,
     Boolean,
     Categorical,
     Date,
@@ -92,6 +93,22 @@ def _set_numpy_to_constructor() -> None:
     }
 
 
+def numpy_values_and_dtype(
+    values: np.ndarray[Any, Any]
+) -> tuple[np.ndarray[Any, Any], type]:
+    """Return numpy values and their associated dtype, adjusting if required."""
+    dtype = values.dtype.type
+    if dtype == np.float16:
+        values = values.astype(np.float32)
+        dtype = values.dtype.type
+    elif dtype == np.datetime64:
+        if np.datetime_data(values.dtype)[0] in DTYPE_TEMPORAL_UNITS:
+            values = values.astype(np.int64)
+        else:
+            dtype = object
+    return values, dtype
+
+
 def numpy_type_to_constructor(dtype: type[np.dtype[Any]]) -> Callable[..., PySeries]:
     """Get the right PySeries constructor for the given Polars dtype."""
     if _NUMPY_TYPE_TO_CONSTRUCTOR is None:

py-polars/polars/internals/construction.py

@@ -17,7 +17,6 @@
 
 from polars import internals as pli
 from polars.datatypes import (
-    DTYPE_TEMPORAL_UNITS,
     Categorical,
     ColumnsType,
     Date,
@@ -37,10 +36,16 @@
 )
 from polars.datatypes_constructor import (
     numpy_type_to_constructor,
+    numpy_values_and_dtype,
     polars_type_to_constructor,
     py_type_to_constructor,
 )
-from polars.dependencies import _NUMPY_AVAILABLE, _PANDAS_TYPE, _PYARROW_AVAILABLE
+from polars.dependencies import (
+    _NUMPY_AVAILABLE,
+    _NUMPY_TYPE,
+    _PANDAS_TYPE,
+    _PYARROW_AVAILABLE,
+)
 from polars.dependencies import numpy as np
 from polars.dependencies import pandas as pd
 from polars.dependencies import pyarrow as pa
@@ -146,32 +151,18 @@ def numpy_to_pyseries(
         values = np.array(values)
 
     if len(values.shape) == 1:
-        dtype = values.dtype.type
-        if dtype == np.float16:
-            values = values.astype(np.float32)
-            dtype = values.dtype.type
-        elif (
-            dtype == np.datetime64
-            and np.datetime_data(values.dtype)[0] not in DTYPE_TEMPORAL_UNITS
-        ):
-            dtype = object
-
+        values, dtype = numpy_values_and_dtype(values)
         constructor = numpy_type_to_constructor(dtype)
-
-        if dtype == np.float32 or dtype == np.float64:
-            return constructor(name, values, nan_to_null)
-        elif dtype == np.datetime64:
-            return constructor(name, values.astype(np.int64), strict)
-        else:
-            return constructor(name, values, strict)
+        return constructor(
+            name, values, nan_to_null if dtype in (np.float32, np.float64) else strict
+        )
     elif len(values.shape) == 2:
         pyseries_container = []
         for row in range(values.shape[0]):
             pyseries_container.append(
                 numpy_to_pyseries("", values[row, :], strict, nan_to_null)
             )
         return PySeries.new_series_list(name, pyseries_container, False)
-
     else:
         return PySeries.new_object(name, values, strict)
 
@@ -250,6 +241,7 @@ def sequence_to_pyseries(
     dtype: PolarsDataType | None = None,
     strict: bool = True,
     dtype_if_empty: PolarsDataType | None = None,
+    nan_to_null: bool = False,
 ) -> PySeries:
     """Construct a PySeries from a sequence."""
     python_dtype: type | None = None
@@ -383,8 +375,20 @@ def sequence_to_pyseries(
             # Convert mixed sequences like `[[12], "foo", 9]`
             return PySeries.new_object(name, values, strict)
 
+        elif (
+            _NUMPY_TYPE(value)
+            and isinstance(value, np.ndarray)
+            and len(value.shape) == 1
+        ):
+            return PySeries.new_series_list(
+                name,
+                [numpy_to_pyseries("", v, strict, nan_to_null) for v in values],
+                strict,
+            )
+
         elif python_dtype == pli.Series:
             return PySeries.new_series_list(name, [v._s for v in values], strict)
+
         elif python_dtype == PySeries:
             return PySeries.new_series_list(name, values, strict)
         else:

py-polars/polars/internals/series/series.py

@@ -255,7 +255,12 @@ def __init__(
             )
         elif isinstance(values, Sequence):
             self._s = sequence_to_pyseries(
-                name, values, dtype=dtype, strict=strict, dtype_if_empty=dtype_if_empty
+                name,
+                values,
+                dtype=dtype,
+                strict=strict,
+                dtype_if_empty=dtype_if_empty,
+                nan_to_null=nan_to_null,
             )
         elif _PANDAS_TYPE(values) and isinstance(values, (pd.Series, pd.DatetimeIndex)):
             self._s = pandas_to_pyseries(name, values)

py-polars/polars/io.py

@@ -1065,7 +1065,7 @@ def read_sql(
 
     Notes
     -----
-    Make sure to install connectorx>=0.2.2. Read the documentation
+    Make sure to install connectorx>=0.3.1. Read the documentation
     `here <https://sfu-db.github.io/connector-x/intro.html>`_.
 
     Examples
@@ -1100,7 +1100,7 @@ def read_sql(
         import connectorx as cx
     except ImportError:
         raise ImportError(
-            "connectorx is not installed. Please run `pip install connectorx>=0.2.2`."
+            "connectorx is not installed. Please run `pip install connectorx>=0.3.1`."
         ) from None
 
     tbl = cx.read_sql(

py-polars/tests/unit/test_series.py

@@ -42,6 +42,7 @@ def test_cum_agg() -> None:
 
 
 def test_init_inputs(monkeypatch: Any) -> None:
+    nan = float("nan")
     for flag in [False, True]:
         monkeypatch.setattr(pl.internals.construction, "_PYARROW_AVAILABLE", flag)
         # Good inputs
@@ -71,11 +72,34 @@ def test_init_inputs(monkeypatch: Any) -> None:
         )
         assert pl.Series("a", [10000, 20000, 30000], dtype=pl.Time).dtype == pl.Time
 
-        # 2d numpy array
-        res = pl.Series(name="a", values=np.array([[1, 2], [3, 4]], dtype=np.int64))
-        assert res.dtype == pl.List(pl.Int64)
-        assert res[0].to_list() == [1, 2]
-        assert res[1].to_list() == [3, 4]
+        # 2d numpy array and/or list of 1d numpy arrays
+        for res in (
+            pl.Series(
+                name="a",
+                values=np.array([[1, 2], [3, nan]], dtype=np.float32),
+                nan_to_null=True,
+            ),
+            pl.Series(
+                name="a",
+                values=[
+                    np.array([1, 2], dtype=np.float32),
+                    np.array([3, nan], dtype=np.float32),
+                ],
+                nan_to_null=True,
+            ),
+            pl.Series(
+                name="a",
+                values=(
+                    np.ndarray((2,), np.float32, np.array([1, 2], dtype=np.float32)),
+                    np.ndarray((2,), np.float32, np.array([3, nan], dtype=np.float32)),
+                ),
+                nan_to_null=True,
+            ),
+        ):
+            assert res.dtype == pl.List(pl.Float32)
+            assert res[0].to_list() == [1.0, 2.0]
+            assert res[1].to_list() == [3.0, None]
+
         assert pl.Series(
             values=np.array([["foo", "bar"], ["foo2", "bar2"]])
         ).dtype == pl.List(pl.Utf8)