Backport PR matplotlib#31707: Fix violinplot crash on empty datasets (matplotlib#31700)

Author: rahulrathnavel

doc/api/next_api_changes/behavior/violinplot_empty.rst

@@ -0,0 +1,4 @@
+Axes.violinplot and cbook.violin_stats ignore non-finite values
+---------------------------------------------------------------
+
+`~matplotlib.axes.Axes.violinplot` and `matplotlib.cbook.violin_stats` now ignore masked and non-finite (NaN and inf) values.

lib/matplotlib/axes/_axes.py

@@ -8893,6 +8893,8 @@ def violinplot(self, dataset, positions=None, vert=None,
             - sequence of 1D arrays: A violin is drawn for each array in the sequence.
             - 2D array: A violin is drawn for each column in the array.
 
+            Non-finite and masked values are ignored.
+
         positions : array-like, default: [1, 2, ..., n]
             The positions of the violins; i.e. coordinates on the x-axis for
             vertical violins (or y-axis for horizontal violins).
@@ -9264,7 +9266,8 @@ def cycle_color(color, alpha=None):
         for stats, pos, width, facecolor in bodies_zip:
             # The 0.5 factor reflects the fact that we plot from v-p to v+p.
             vals = np.array(stats['vals'])
-            vals = 0.5 * width * vals / vals.max()
+            if len(vals) > 0:
+                vals = 0.5 * width * vals / vals.max()
             bodies += [fill(stats['coords'],
                             -vals + pos if side in ['both', 'low'] else pos,
                             vals + pos if side in ['both', 'high'] else pos,

lib/matplotlib/cbook.py

@@ -1499,7 +1499,8 @@ def violin_stats(X, method=("GaussianKDE", "scott"), points=100, quantiles=None)
     ----------
     X : 1D array or sequence of 1D arrays or 2D array
         Sample data that will be used to produce the gaussian kernel density
-        estimates. Possible values:
+        estimates. Non-finite and masked values are ignored.
+        Possible values:
 
         - 1D array: Statistics are computed for that array.
         - sequence of 1D arrays: Statistics are computed for each array in the sequence.
@@ -1586,29 +1587,34 @@ def _kde_method(x, coords):
                          " must have the same length")
 
     # Zip x and quantiles
-    for (x, q) in zip(X, quantiles):
-        # Dictionary of results for this distribution
-        stats = {}
-
-        # Calculate basic stats for the distribution
-        min_val = np.min(x)
-        max_val = np.max(x)
-        quantile_val = np.percentile(x, 100 * q)
+    for (x, quantile) in zip(X, quantiles):
+        x = np.asarray(x)
+        x, = delete_masked_points(x)
 
-        # Evaluate the kernel density estimate
-        coords = np.linspace(min_val, max_val, points)
-        stats['vals'] = method(x, coords)
-        stats['coords'] = coords
-
-        # Store additional statistics for this distribution
-        stats['mean'] = np.mean(x)
-        stats['median'] = np.median(x)
-        stats['min'] = min_val
-        stats['max'] = max_val
-        stats['quantiles'] = np.atleast_1d(quantile_val)
-
-        # Append to output
-        vpstats.append(stats)
+        if len(x) == 0:
+            vpstats.append({
+                'vals': np.array([]),
+                'coords': np.array([]),
+                'mean': np.nan,
+                'median': np.nan,
+                'min': np.nan,
+                'max': np.nan,
+                'quantiles': np.array([]),
+            })
+        else:
+            min_val = np.min(x)
+            max_val = np.max(x)
+            coords = np.linspace(min_val, max_val, points)
+
+            vpstats.append({
+                'vals': method(x, coords),
+                'coords': coords,
+                'mean': np.mean(x),
+                'median': np.median(x),
+                'min': min_val,
+                'max': max_val,
+                'quantiles': np.atleast_1d(np.percentile(x, 100 * quantile))
+            })
 
     return vpstats
 

lib/matplotlib/tests/test_axes.py

@@ -10360,3 +10360,10 @@ def test_errorbar_uses_rcparams():
     assert_allclose([cap.get_markeredgewidth() for cap in caplines], 2.5)
     for barcol in barlinecols:
         assert_allclose(barcol.get_linewidths(), 1.75)
+
+
+def test_violinplot_empty_dataset():
+    fig, ax = plt.subplots()
+    # This should not raise an exception
+    parts = ax.violinplot([np.random.randn(100), [], [np.nan, np.nan]])
+    assert len(parts["bodies"]) == 3