Merge pull request #1381 from mrocklin/get-numeric-data

Implement DataFrame._get_numeric_data

CHANGELOG.md

@@ -6,6 +6,7 @@
 - PR #1292 Implemented Bitwise binary ops AND, OR, XOR (&, |, ^)
 - PR #1235 Add GPU-accelerated Parquet Reader
 - PR #1335 Added local_dict arg in `DataFrame.query()`.
+- PR #1381 Add DataFrame._get_numeric_data
 
 ## Improvements
 
@@ -29,7 +30,7 @@
 - PR #1254 CSV Reader: fix data type detection for floating-point numbers in scientific notation
 - PR #1289 Fix looping over each value instead of each category in concatenation
 - PR #1293 Fix Inaccurate error message in join.pyx
-- PR #1308 Add atomicCAS overload for `int8_t`, `int16_t` 
+- PR #1308 Add atomicCAS overload for `int8_t`, `int16_t`
 - PR #1317 Fix catch polymorphic exception by reference in ipc.cu
 - PR #1325 Fix dtype of null bitmasks to int8
 - PR #1326 Update build documentation to use -DCMAKE_CXX11_ABI=ON

python/cudf/dataframe/dataframe.py

@@ -305,6 +305,13 @@ def __array_ufunc__(self, ufunc, method, *inputs, **kwargs):
     def empty(self):
         return not len(self)
 
+    def _get_numeric_data(self):
+        """ Return a dataframe with only numeric data types """
+        columns = [c for c, dt in self.dtypes.items()
+                   if dt != object and
+                   not pd.api.types.is_categorical_dtype(dt)]
+        return self[columns]
+
     def assign(self, **kwargs):
         """
         Assign columns to DataFrame from keyword arguments.

python/cudf/tests/test_dataframe.py

@@ -2050,3 +2050,14 @@ def test_series_to_gpu_array(nan_value):
     s = Series([0, 1, None, 3])
     np.testing.assert_array_equal(s.to_array(nan_value),
                                   s.to_gpu_array(nan_value).copy_to_host())
+
+
+def test_get_numeric_data():
+    pdf = pd.DataFrame({
+        'x': [1, 2, 3],
+        'y': [1., 2., 3.],
+        'z': ['a', 'b', 'c']
+    })
+    gdf = gd.from_pandas(pdf)
+
+    assert_eq(pdf._get_numeric_data(), gdf._get_numeric_data())