ENH: add basic DataFrame.from_arrow class method for importing through Arrow PyCapsule interface

Author: jorisvandenbossche

pandas/core/frame.py

@@ -1746,6 +1746,41 @@ def __rmatmul__(self, other) -> DataFrame:
     # ----------------------------------------------------------------------
     # IO methods (to / from other formats)
 
+    @classmethod
+    def from_arrow(cls, data):
+        """
+        Construct a DataFrame from a tabular Arrow object.
+
+        This function accepts any tabular Arrow object implementing
+        the `Arrow PyCapsule Protocol`_ (i.e. having an ``__arrow_c_array__``
+        or ``__arrow_c_stream__`` method).
+
+        This function currently relies on ``pyarrow`` to convert the tabular
+        object in Arrow format to pandas.
+
+        .. _Arrow PyCapsule Protocol: https://arrow.apache.org/docs/format/CDataInterface/PyCapsuleInterface.html
+
+        .. versionadded:: 3.0
+
+        Parameters
+        ----------
+        data : pyarrow.Table or Arrow-compatible table
+            Any tabular object implementing the Arrow PyCapsule Protocol
+            (i.e. has an ``__arrow_c_array__`` or ``__arrow_c_stream__``
+            method).
+
+        Returns
+        -------
+        DataFrame
+
+        """
+        pa = import_optional_dependency("pyarrow", min_version="14.0.0")
+        if not isinstance(data, pa.Table):
+            data = pa.table(data)
+
+        df = data.to_pandas()
+        return df
+
     @classmethod
     def from_dict(
         cls,

pandas/tests/frame/test_arrow_interface.py

@@ -7,6 +7,7 @@
 import pandas.util._test_decorators as td
 
 import pandas as pd
+import pandas._testing as tm
 
 pa = pytest.importorskip("pyarrow")
 
@@ -47,3 +48,42 @@ def test_dataframe_to_arrow():
     table = pa.RecordBatchReader.from_stream(df, schema=schema).read_all()
     expected = expected.cast(schema)
     assert table.equals(expected)
+
+
+class ArrowArrayWrapper:
+    def __init__(self, batch):
+        self.array = batch
+
+    def __arrow_c_array__(self, requested_schema=None):
+        return self.array.__arrow_c_array__(requested_schema)
+
+
+class ArrowStreamWrapper:
+    def __init__(self, table):
+        self.stream = table
+
+    def __arrow_c_stream__(self, requested_schema=None):
+        return self.stream.__arrow_c_stream__(requested_schema)
+
+
+@td.skip_if_no("pyarrow", min_version="14.0")
+def test_dataframe_from_arrow():
+    # objects with __arrow_c_stream__
+    table = pa.table({"a": [1, 2, 3], "b": ["a", "b", "c"]})
+
+    result = pd.DataFrame.from_arrow(table)
+    expected = pd.DataFrame({"a": [1, 2, 3], "b": ["a", "b", "c"]})
+    tm.assert_frame_equal(result, expected)
+
+    # not only pyarrow object are supported
+    result = pd.DataFrame.from_arrow(ArrowStreamWrapper(table))
+    tm.assert_frame_equal(result, expected)
+
+    # objects with __arrow_c_array__
+    batch = pa.record_batch([[1, 2, 3], ["a", "b", "c"]], names=["a", "b"])
+
+    result = pd.DataFrame.from_arrow(table)
+    tm.assert_frame_equal(result, expected)
+
+    result = pd.DataFrame.from_arrow(ArrowArrayWrapper(batch))
+    tm.assert_frame_equal(result, expected)