FIX-#7157: Make sure quantile function works with numeric_only=True (#7160)

Co-authored-by: Iaroslav Igoshev <[email protected]>
Signed-off-by: Anatoly Myachev <[email protected]>

Author: anmyachev

modin/core/storage_formats/pandas/query_compiler.py

@@ -2588,7 +2588,7 @@ def quantile_for_list_of_values(self, **kwargs):
         axis = kwargs.get("axis", 0)
         q = kwargs.get("q")
         numeric_only = kwargs.get("numeric_only", True)
-        assert isinstance(q, (pandas.Series, np.ndarray, pandas.Index, list))
+        assert isinstance(q, (pandas.Series, np.ndarray, pandas.Index, list, tuple))
 
         if numeric_only:
             new_columns = self._modin_frame.numeric_columns()

modin/pandas/base.py

@@ -2304,6 +2304,7 @@ def quantile(
         def check_dtype(t):
             return is_numeric_dtype(t) or lib.is_np_dtype(t, "mM")
 
+        numeric_only_df = self
         if not numeric_only:
             # If not numeric_only and columns, then check all columns are either
             # numeric, timestamp, or timedelta
@@ -2322,31 +2323,33 @@ def check_dtype(t):
                             )
                         )
         else:
-            # Normally pandas returns this near the end of the quantile, but we
-            # can't afford the overhead of running the entire operation before
-            # we error.
-            if not any(is_numeric_dtype(t) for t in self._get_dtypes()):
-                raise ValueError("need at least one array to concatenate")
+            numeric_only_df = self.drop(
+                columns=[
+                    i for i in self.dtypes.index if not is_numeric_dtype(self.dtypes[i])
+                ]
+            )
 
         # check that all qs are between 0 and 1
         validate_percentile(q)
-        axis = self._get_axis_number(axis)
-        if isinstance(q, (pandas.Series, np.ndarray, pandas.Index, list)):
-            return self.__constructor__(
-                query_compiler=self._query_compiler.quantile_for_list_of_values(
+        axis = numeric_only_df._get_axis_number(axis)
+        if isinstance(q, (pandas.Series, np.ndarray, pandas.Index, list, tuple)):
+            return numeric_only_df.__constructor__(
+                query_compiler=numeric_only_df._query_compiler.quantile_for_list_of_values(
                     q=q,
                     axis=axis,
-                    numeric_only=numeric_only,
+                    # `numeric_only=True` has already been processed by using `self.drop` function
+                    numeric_only=False,
                     interpolation=interpolation,
                     method=method,
                 )
             )
         else:
-            result = self._reduce_dimension(
-                self._query_compiler.quantile_for_single_value(
+            result = numeric_only_df._reduce_dimension(
+                numeric_only_df._query_compiler.quantile_for_single_value(
                     q=q,
                     axis=axis,
-                    numeric_only=numeric_only,
+                    # `numeric_only=True` has already been processed by using `self.drop` function
+                    numeric_only=False,
                     interpolation=interpolation,
                     method=method,
                 )

modin/tests/pandas/dataframe/test_window.py

@@ -669,6 +669,32 @@ def test_quantile(request, data, q):
             modin_df.T.quantile(q)
 
 
+def test_quantile_7157():
+    # for details: https://github.com/modin-project/modin/issues/7157
+    n_rows = 100
+    n_fcols = 10
+    n_mcols = 5
+
+    df1_md, df1_pd = create_test_dfs(
+        random_state.rand(n_rows, n_fcols),
+        columns=[f"feat_{i}" for i in range(n_fcols)],
+    )
+    df2_md, df2_pd = create_test_dfs(
+        {
+            "test_string1": ["test_string2" for _ in range(n_rows)]
+            for _ in range(n_mcols)
+        }
+    )
+    df3_md = pd.concat([df2_md, df1_md], axis=1)
+    df3_pd = pandas.concat([df2_pd, df1_pd], axis=1)
+
+    eval_general(df3_md, df3_pd, lambda df: df.quantile(0.25, numeric_only=True))
+    eval_general(df3_md, df3_pd, lambda df: df.quantile((0.25,), numeric_only=True))
+    eval_general(
+        df3_md, df3_pd, lambda df: df.quantile((0.25, 0.75), numeric_only=True)
+    )
+
+
 @pytest.mark.parametrize("axis", ["rows", "columns"])
 @pytest.mark.parametrize(
     "na_option", ["keep", "top", "bottom"], ids=["keep", "top", "bottom"]