Raise ValueError instead of coercing non-numeric data to numeric types

Author: Brendan Drury

doc/source/whatsnew/v1.4.0.rst

@@ -485,7 +485,7 @@ Period
 
 Plotting
 ^^^^^^^^
-- Fixed bug in :meth:`boxplot._grouped_plot_by_column` where trying to plot data of dtype ``object`` caused plotting to fail even if some data was numeric
+- When given non-numeric data, :meth:`DataFrame.boxplot` now raises a ``ValueError`` rather than a cryptic ``KeyError`` or ``ZeroDivsionError``, in line with other plotting functions like :meth:`DataFrame.hist`. (:issue:`43480`)
 -
 
 Groupby/resample/rolling

pandas/plotting/_matplotlib/boxplot.py

@@ -390,8 +390,12 @@ def plot_group(keys, values, ax: Axes):
             rc = {"figure.figsize": figsize} if figsize is not None else {}
             with plt.rc_context(rc):
                 ax = plt.gca()
-        data = data.apply(pd.to_numeric, errors="ignore")
         data = data._get_numeric_data()
+        naxes = len(data.columns)
+        if naxes == 0:
+            raise ValueError(
+                "boxplot method requires numerical columns, nothing to plot."
+            )
         if columns is None:
             columns = data.columns
         else:

pandas/tests/plotting/test_boxplot_method.py

@@ -547,7 +547,9 @@ def test_groupby_boxplot_object(self):
         # GH 43480
         df = self.hist_df.astype("object")
         grouped = df.groupby("gender")
-        _check_plot_works(grouped.boxplot, subplots=False)
+        msg = "boxplot method requires numerical columns, nothing to plot"
+        with pytest.raises(ValueError, match=msg):
+            _check_plot_works(grouped.boxplot, subplots=False)
 
     def test_boxplot_multiindex_column(self):
         # GH 16748