[python][RDF] Handle the creation of an empty RDataFrame in _MakeNumpyDataFrame

bindings/pyroot/pythonizations/python/ROOT/_pythonization/_rdataframe.py

@@ -203,7 +203,7 @@ def pypowarray(numpyvec, pow):
 
 \endpythondoc
 '''
-import sys
+
 from . import pythonization
 from ._pyz_utils import MethodTemplateGetter, MethodTemplateWrapper
 
@@ -246,18 +246,18 @@ def RDataFrameAsNumpy(df, columns=None, exclude=None, lazy=False):
 
     # Early check for numpy
     try:
-        import numpy
-    except:
+        import numpy  # noqa: F401
+    except ImportError:
         raise ImportError("Failed to import numpy during call of RDataFrame.AsNumpy.")
 
     # Find all column names in the dataframe if no column are specified
     if not columns:
         columns = [str(c) for c in df.GetColumnNames()]
 
     # Exclude the specified columns
-    if exclude == None:
+    if exclude is None:
         exclude = []
-    columns = [col for col in columns if not col in exclude]
+    columns = [col for col in columns if col not in exclude]
 
     # Register Take action for each column
     result_ptrs = {}
@@ -321,6 +321,7 @@ def GetValue(self):
 
         if self._py_arrays is None:
             import numpy
+
             from ROOT._pythonization._rdf_utils import ndarray
 
             # Convert the C++ vectors to numpy arrays
@@ -474,7 +475,7 @@ def pythonize_rdataframe(klass):
 
     klass._OriginalFilter = klass.Filter
     klass._OriginalDefine = klass.Define
-    from ._rdf_pyz import _PyFilter, _PyDefine
+    from ._rdf_pyz import _PyDefine, _PyFilter
 
     klass.Filter = _PyFilter
     klass.Define = _PyDefine
@@ -487,10 +488,16 @@ def _make_name_rvec_pair(key, value):
     if not isinstance(key, str):
         raise RuntimeError("Object not convertible: Dictionary key is not convertible to a string.")
 
-    # Convert value to RVec and attach to dictionary
-    pyvec = ROOT.VecOps.AsRVec(value)
-    if not pyvec:
-        raise RuntimeError("Object not convertible: Dictionary entry " + key + " is not convertible with AsRVec.")
+    try:
+        # Convert value to RVec and attach to dictionary
+        pyvec = ROOT.VecOps.AsRVec(value)
+    except TypeError as e:
+        if "Cannot create an RVec from a numpy array of data type object" in str(e):
+            raise RuntimeError(
+                f"Failure in creating column '{key}' for RDataFrame: the input column type is 'object', which is not supported. Make sure your column type is supported."
+            ) from e
+        else:
+            raise
 
     # Add pairs of column name and associated RVec to signature
     return ROOT.std.pair["std::string", type(pyvec)](key, ROOT.std.move(pyvec))

bindings/pyroot/pythonizations/test/rdataframe_makenumpy.py

@@ -36,6 +36,26 @@ def test_object_dtype_strings(self):
         entries = df.AsNumpy()["x"]
         self.assertTrue(np.array_equal(entries, np.array(["test_string_1", "test_string_2"], dtype=object)))
 
+    def test_empty_arrays(self):
+        """
+        Test creating an RDataFrame from an empty numpy array with different data types
+        """
+        for dtype in self.dtypes:
+            data = {"x": np.array([], dtype=dtype)}
+            df = ROOT.RDF.FromNumpy(data)
+            colnames = df.GetColumnNames()
+            self.assertIn("x", colnames)
+            self.assertEqual(df.Count().GetValue(), 0)
+
+        data_obj = {"x": np.array([], dtype="object")}
+        with self.assertRaises(RuntimeError) as context:
+            df = ROOT.RDF.FromNumpy(data_obj)
+
+        self.assertIn(
+            "Failure in creating column 'x' for RDataFrame: the input column type is 'object', which is not supported. Make sure your column type is supported",
+            str(context.exception),
+        )
+
     def test_multiple_columns(self):
         """
         Test reading multiple columns