Improve robustness to empty data in Plot and Nominal (#3202)

* Improve robustness to missing data in Plot and Nominal

* Fix Plot.on tests to work with new logic

Author: mwaskom

doc/whatsnew/v0.12.2.rst

@@ -8,6 +8,8 @@ v0.12.2 (Unreleased)
 
 - |Fix| Fixed a bug where legends for numeric variables with large values with be incorrectly shown (i.e. with a missing offset or exponent; :pr:`3187`).
 
+- |Fix| Improve robustness to empty data in several components of the objects interface (:pr:`3202`).
+
 - |Fix| Fixed a regression in v0.12.0 where manually-added labels could have duplicate legend entries (:pr:`3116`).
 
 - |Fix| Fixed a bug in :func:`histplot` with `kde=True` and `log_scale=True` where the curve was not scaled properly (:pr:`3173`).

seaborn/_core/plot.py

@@ -1466,8 +1466,6 @@ def _setup_split_generator(
         self, grouping_vars: list[str], df: DataFrame, subplots: list[dict[str, Any]],
     ) -> Callable[[], Generator]:
 
-        allow_empty = False  # TODO will need to recreate previous categorical plots
-
         grouping_keys = []
         grouping_vars = [
             v for v in grouping_vars if v in df and v not in ["col", "row"]
@@ -1506,7 +1504,8 @@ def split_generator(keep_na=False) -> Generator:
                         subplot_keys[dim] = view[dim]
 
                 if not grouping_vars or not any(grouping_keys):
-                    yield subplot_keys, axes_df.copy(), view["ax"]
+                    if not axes_df.empty:
+                        yield subplot_keys, axes_df.copy(), view["ax"]
                     continue
 
                 grouped_df = axes_df.groupby(grouping_vars, sort=False, as_index=False)
@@ -1526,7 +1525,7 @@ def split_generator(keep_na=False) -> Generator:
                         # case this option could be removed
                         df_subset = axes_df.loc[[]]
 
-                    if df_subset.empty and not allow_empty:
+                    if df_subset.empty:
                         continue
 
                     sub_vars = dict(zip(grouping_vars, key))

seaborn/_core/scales.py

@@ -163,7 +163,7 @@ def _setup(
             new = new.label()
 
         # TODO flexibility over format() which isn't great for numbers / dates
-        stringify = np.vectorize(format)
+        stringify = np.vectorize(format, otypes=["object"])
 
         units_seed = categorical_order(data, new.order)
 

tests/_core/test_plot.py

@@ -680,8 +680,9 @@ def test_matplotlib_object_creation(self):
     def test_empty(self):
 
         m = MockMark()
-        Plot().plot()
+        Plot().add(m).plot()
         assert m.n_splits == 0
+        assert not m.passed_data
 
     def test_no_orient_variance(self):
 
@@ -1086,7 +1087,7 @@ def test_on_axes(self):
 
         ax = mpl.figure.Figure().subplots()
         m = MockMark()
-        p = Plot().on(ax).add(m).plot()
+        p = Plot([1], [2]).on(ax).add(m).plot()
         assert m.passed_axes == [ax]
         assert p._figure is ax.figure
 
@@ -1095,7 +1096,7 @@ def test_on_figure(self, facet):
 
         f = mpl.figure.Figure()
         m = MockMark()
-        p = Plot().on(f).add(m)
+        p = Plot([1, 2], [3, 4]).on(f).add(m)
         if facet:
             p = p.facet(["a", "b"])
         p = p.plot()
@@ -1112,7 +1113,7 @@ def test_on_subfigure(self, facet):
         sf1, sf2 = mpl.figure.Figure().subfigures(2)
         sf1.subplots()
         m = MockMark()
-        p = Plot().on(sf2).add(m)
+        p = Plot([1, 2], [3, 4]).on(sf2).add(m)
         if facet:
             p = p.facet(["a", "b"])
         p = p.plot()

tests/_core/test_scales.py

@@ -555,6 +555,12 @@ class MockProperty(IntervalProperty):
         s = Nominal((2, 4))._setup(x, MockProperty())
         assert_array_equal(s(x), [4, np.sqrt(10), 2, np.sqrt(10)])
 
+    def test_empty_data(self):
+
+        x = pd.Series([], dtype=object, name="x")
+        s = Nominal()._setup(x, Coordinate())
+        assert_array_equal(s(x), [])
+
 
 class TestTemporal: