Use histogram for comparison plots

app.py

@@ -12,7 +12,6 @@ import seaborn as sns
 import matplotlib.pyplot as plt
 from datasets import load_dataset
 from scipy.special import expit
-# from matplotlib.ticker import MultipleLocator
 
 from hdinterval import HDI, HDInterval
 
@@ -133,35 +132,50 @@ class RankPlotter(DataPlotter):
         ax.set_yticks(self.y, self.df['model'])
 
 class ComparisonPlotter(DataPlotter):
+    _uncertain = 0.5
+
+    @staticmethod
+    def to_relative(hdi, ax):
+        (lhs, rhs) = ax.get_xlim()
+        length = rhs - lhs
+        yield from (abs(lhs - x) / length for x in hdi)
+
     def __init__(self, df, model_1, model_2, ci):
         super().__init__(compare(df, model_1, model_2))
-        self.hdi = HDInterval(self.df)
+        self.interval = HDInterval(self.df)
         self.ci = ci
 
     def draw(self, ax):
-        interval = self.hdi(self.ci)
+        hdi = self.interval(self.ci)
 
-        sns.ecdfplot(self.df, ax=ax)
+        ax = sns.histplot(self.df, stat='density')
+
+        top = max(x.get_height() for x in ax.patches)
+        y = top * 1.05
 
         (_, color, *_) = sns.color_palette()
-        ax.axvline(x=self.df.median(),
-                   color=color,
-                   linestyle='dashed')
-        ax.axvspan(xmin=interval.lower,
-                   xmax=interval.upper,
-                   alpha=0.15,
+        (xmin, xmax) = self.to_relative(hdi, ax)
+        linestyle = 'dashed' if self._uncertain in hdi else 'solid'
+        ax.axhline(y=y,
+                   xmin=xmin,
+                   xmax=xmax,
+                   linestyle=linestyle,
                    color=color)
         ax.set_xlabel('Pr(M$_{1}$ \u003E M$_{2}$)')
-        # for i in ('x', 'y'):
-        #     lim = getattr(ax, f'set_{i}lim')
-        #     lim(-0.01, 1.01)
-        # ax.xaxis.set_major_locator(MultipleLocator(base=0.1, offset=0))
+
+        x = (hdi.lower + hdi.upper) / 2
+        ax.text(x=x,
+                y=y,
+                s=f'{self.ci:.0%} HDI',
+                backgroundcolor='white',
+                horizontalalignment='center',
+                verticalalignment='center')
 
         try:
-            ci_mid = self.hdi.at(0.5)
+            ci_min = self.interval.at(self._uncertain)
             ax.text(x=0.01,
                     y=0.99,
-                    s=f'0.5-min HDI: {ci_mid:.0%}',
+                    s=f'0.5 \u2248\u2208 {ci_min:.0%} HDI',
                     horizontalalignment='left',
                     verticalalignment='top',
                     transform=ax.transAxes)
@@ -177,10 +191,11 @@ class ComparisonMenu:
         self.ci = ci
 
     def __call__(self, model_1, model_2, ci):
-        ci /= 100
-        cp = ComparisonPlotter(self.df, model_1, model_2, ci)
+        if model_1 and model_2:
+            ci /= 100
+            cp = ComparisonPlotter(self.df, model_1, model_2, ci)
 
-        return cp.plot()
+            return cp.plot()
 
     def build_and_get(self):
         models = self.df['model'].unique()
@@ -246,13 +261,14 @@ def layout(tab):
                 gr.Markdown('''
 
                 Probability that Model 1 is preferred to Model 2. The
-                solid blue curve is a CDF of that distribution;
-                formally the inverse logit of the difference in model
-                abilities. The dashed orange vertical line is the
-                median, while the band surrounding it is the [highest
-                density
-                interval](https://cran.r-project.org/package=HDInterval)
-                of your choice (default 95%).
+                histogram is represents the distribution of inverse
+                logit of the difference in model abilities. The
+                horizontal line above the histogram marks the chosen
+                [highest density
+                interval](https://cran.r-project.org/package=HDInterval). The
+                line is dashed if the interval overlaps 0.5, solid
+                otherwise. The HDI in the upper left denotes the
+                smallest approximate HDI that is inclusive of 0.5.
 
                 ''')
             with gr.Column():