Fix histplot/kdeplot normalization with weights

doc/releases/v0.12.0.txt

@@ -51,10 +51,12 @@ Other updates
 
 - |Enhancement| Example datasets are now stored in an OS-specific cache location (as determined by `appdirs`) rather than in the user's home directory. Users should feel free to remove `~/seaborn-data` if desired (:pr:`2773`).
 
-- |Fix| FacetGrid subplot titles will no longer be reset when calling :meth:`FacetGrid.map` or :meth:`FacetGrid.map_dataframe` after :meth:`FacetGrid.set_titles` (:pr:`2705`).
-
 - |Fix| Fixed a regression in 0.11.2 that caused some functions to stall indefinitely or raise when the input data had a duplicate index (:pr:`2776`).
 
+- |Fix| Fixed a bug in :func:`histplot` and :func:`kdeplot` where weights were not factored into the normalization (:pr:`2812`).
+
+- |Fix| FacetGrid subplot titles will no longer be reset when calling :meth:`FacetGrid.map` or :meth:`FacetGrid.map_dataframe` after :meth:`FacetGrid.set_titles` (:pr:`2705`).
+
 - |Fix| In :func:`lineplot`, allowed the `dashes` keyword to set the style of a line without mapping a `style` variable (:pr:`2449`).
 
 - |Dependencies| Made `scipy` an optional dependency and added `pip install seaborn[all]` as a method for ensuring the availability of compatible `scipy` and `statsmodels` libraries at install time. This has a few minor implications for existing code, which are explained in the Github pull request (:pr:`2398`).

seaborn/distributions.py

@@ -304,15 +304,19 @@ def _compute_univariate_density(
         # Initialize the estimator object
         estimator = KDE(**estimate_kws)
 
-        all_data = self.plot_data.dropna()
-
         if set(self.variables) - {"x", "y"}:
             if common_grid:
                 all_observations = self.comp_data.dropna()
                 estimator.define_support(all_observations[data_variable])
         else:
             common_norm = False
 
+        all_data = self.plot_data.dropna()
+        if common_norm and "weights" in all_data:
+            whole_weight = all_data["weights"].sum()
+        else:
+            whole_weight = len(all_data)
+
         densities = {}
 
         for sub_vars, sub_data in self.iter_data("hue", from_comp_data=True):
@@ -333,8 +337,10 @@ def _compute_univariate_density(
             # Extract the weights for this subset of observations
             if "weights" in self.variables:
                 weights = sub_data["weights"]
+                part_weight = weights.sum()
             else:
                 weights = None
+                part_weight = len(sub_data)
 
             # Estimate the density of observations at this level
             density, support = estimator(observations, weights=weights)
@@ -344,7 +350,7 @@ def _compute_univariate_density(
 
             # Apply a scaling factor so that the integral over all subsets is 1
             if common_norm:
-                density *= len(sub_data) / len(all_data)
+                density *= part_weight / whole_weight
 
             # Store the density for this level
             key = tuple(sub_vars.items())
@@ -408,21 +414,22 @@ def plot_univariate_histogram(
         histograms = {}
 
         # Do pre-compute housekeeping related to multiple groups
-        # TODO best way to account for facet/semantic?
-        if set(self.variables) - {"x", "y"}:
-
-            all_data = self.comp_data.dropna()
+        all_data = self.comp_data.dropna()
+        all_weights = all_data.get("weights", None)
 
+        if set(self.variables) - {"x", "y"}:  # Check if we'll have multiple histograms
             if common_bins:
-                all_observations = all_data[self.data_variable]
                 estimator.define_bin_params(
-                    all_observations,
-                    weights=all_data.get("weights", None),
+                    all_data[self.data_variable], weights=all_weights
                 )
-
         else:
             common_norm = False
 
+        if common_norm and all_weights is not None:
+            whole_weight = all_weights.sum()
+        else:
+            whole_weight = len(all_data)
+
         # Estimate the smoothed kernel densities, for use later
         if kde:
             # TODO alternatively, clip at min/max bins?
@@ -447,8 +454,10 @@ def plot_univariate_histogram(
 
             if "weights" in self.variables:
                 weights = sub_data["weights"]
+                part_weight = weights.sum()
             else:
                 weights = None
+                part_weight = len(sub_data)
 
             # Do the histogram computation
             heights, edges = estimator(observations, weights=weights)
@@ -478,7 +487,7 @@ def plot_univariate_histogram(
 
             # Apply scaling to normalize across groups
             if common_norm:
-                hist *= len(sub_data) / len(all_data)
+                hist *= part_weight / whole_weight
 
             # Store the finalized histogram data for future plotting
             histograms[key] = hist

seaborn/tests/test_distributions.py

@@ -823,6 +823,19 @@ def test_weights(self):
 
         assert y1 == pytest.approx(2 * y2)
 
+    def test_weight_norm(self, rng):
+
+        vals = rng.normal(0, 1, 50)
+        x = np.concatenate([vals, vals])
+        w = np.repeat([1, 2], 50)
+        ax = kdeplot(x=x, weights=w, hue=w, common_norm=True)
+
+        # Recall that artists are added in reverse of hue order
+        x1, y1 = ax.lines[0].get_xydata().T
+        x2, y2 = ax.lines[1].get_xydata().T
+
+        assert integrate(y1, x1) == pytest.approx(2 * integrate(y2, x2))
+
     def test_sticky_edges(self, long_df):
 
         f, (ax1, ax2) = plt.subplots(ncols=2)
@@ -1397,6 +1410,21 @@ def test_weights_with_missing(self, missing_df):
         total_weight = missing_df[["x", "s"]].dropna()["s"].sum()
         assert sum(bar_heights) == pytest.approx(total_weight)
 
+    def test_weight_norm(self, rng):
+
+        vals = rng.normal(0, 1, 50)
+        x = np.concatenate([vals, vals])
+        w = np.repeat([1, 2], 50)
+        ax = histplot(
+            x=x, weights=w, hue=w, common_norm=True, stat="density", bins=5
+        )
+
+        # Recall that artists are added in reverse of hue order
+        y1 = [bar.get_height() for bar in ax.patches[:5]]
+        y2 = [bar.get_height() for bar in ax.patches[5:]]
+
+        assert sum(y1) == 2 * sum(y2)
+
     def test_discrete(self, long_df):
 
         ax = histplot(long_df, x="s", discrete=True)