Point stat choice

src/beanmachine/ppl/diagnostics/tools/js/src/marginal1d/callbacks.ts

@@ -6,14 +6,13 @@
  */
 
 import {Axis} from '@bokehjs/models/axes/axis';
-import {cumulativeSum} from '../stats/array';
+import {arrayMean, arrayMedian, cumulativeSum} from '../stats/array';
 import {scaleToOne} from '../stats/dataTransformation';
 import {
   interval as hdiInterval,
   data as hdiData,
 } from '../stats/highestDensityInterval';
 import {oneD} from '../stats/marginal';
-import {mean as computeMean} from '../stats/pointStatistic';
 import {interpolatePoints} from '../stats/utils';
 import * as interfaces from './interfaces';
 
@@ -46,6 +45,8 @@ export const updateAxisLabel = (axis: Axis, label: string | null): void => {
  * @param {number[]} marginalX - The support of the Kernel Density Estimate of the
  *     random variable.
  * @param {number[]} marginalY - The Kernel Density Estimate of the random variable.
+ * @param {number} activeStatistic - The statistic to show in the tool. 0 is the mean
+ *     and 1 is the median.
  * @param {number | null} [hdiProb=null] - The highest density interval probability
  *     value. If the default value is not overwritten, then the default HDI probability
  *     is 0.89. See Statistical Rethinking by McElreath for a description as to why this
@@ -62,6 +63,7 @@ export const computeStats = (
   rawData: number[],
   marginalX: number[],
   marginalY: number[],
+  activeStatistic: number,
   hdiProb: number | null = null,
   text_align: string[] = ['right', 'center', 'left'],
   x_offset: number[] = [-5, 0, 5],
@@ -72,24 +74,44 @@ export const computeStats = (
 
   // Compute the point statistics for the KDE, and create labels to display them in the
   // figures.
-  const mean = computeMean(rawData);
+  const mean = arrayMean(rawData);
+  const median = arrayMedian(rawData);
   const hdiBounds = hdiInterval(rawData, hdiProbability);
-  const x = [hdiBounds.lowerBound, mean, hdiBounds.upperBound];
-  const y = interpolatePoints({x: marginalX, y: marginalY, points: x});
-  const text = [
+  let x = [hdiBounds.lowerBound, mean, median, hdiBounds.upperBound];
+  let y = interpolatePoints({x: marginalX, y: marginalY, points: x});
+  let text = [
     `Lower HDI: ${hdiBounds.lowerBound.toFixed(3)}`,
     `Mean: ${mean.toFixed(3)}`,
+    `Median: ${median.toFixed(3)}`,
     `Upper HDI: ${hdiBounds.upperBound.toFixed(3)}`,
   ];
 
-  return {
+  // We will filter the output based on the active statistic from the tool.
+  let mask: number[] = [];
+  if (activeStatistic === 0) {
+    mask = [0, 1, 3];
+  } else if (activeStatistic === 1) {
+    mask = [0, 2, 3];
+  }
+  x = mask.map((i) => {
+    return x[i];
+  });
+  y = mask.map((i) => {
+    return y[i];
+  });
+  text = mask.map((i) => {
+    return text[i];
+  });
+
+  const output = {
     x: x,
     y: y,
     text: text,
     text_align: text_align,
     x_offset: x_offset,
     y_offset: y_offset,
   };
+  return output;
 };
 
 /**
@@ -100,6 +122,8 @@ export const computeStats = (
  *     calculating the Kernel Density Estimate (KDE).
  * @param {number} hdiProbability - The highest density interval probability to use when
  *     calculating the HDI.
+ * @param {number} activeStatistic - The statistic to show in the tool. 0 is the mean
+ *     and 1 is the median.
  * @returns {interfaces.Data} The marginal distribution and cumulative
  *     distribution calculated from the given random variable data. Point statistics are
  *     also calculated.
@@ -108,6 +132,7 @@ export const computeData = (
   data: number[],
   bwFactor: number,
   hdiProbability: number,
+  activeStatistic: number,
 ): interfaces.Data => {
   const output = {} as interfaces.Data;
   for (let i = 0; i < figureNames.length; i += 1) {
@@ -125,7 +150,13 @@ export const computeData = (
     }
 
     // Compute the point statistics for the given data.
-    const stats = computeStats(data, distribution.x, distribution.y, hdiProbability);
+    const stats = computeStats(
+      data,
+      distribution.x,
+      distribution.y,
+      activeStatistic,
+      hdiProbability,
+    );
 
     output[figureName] = {
       distribution: distribution,
@@ -150,6 +181,7 @@ export const computeData = (
  *     application.
  * @param {interfaces.Figures} figures - Bokeh figures shown in the application.
  * @param {interfaces.Tooltips} tooltips - Bokeh tooltips shown on the glyphs.
+ * @param {interfaces.Widgets} widgets - Bokeh widget object for the tool.
  * @returns {number} We display the value of the bandwidth used for computing the Kernel
  *     Density Estimate in a div, and must return that value here in order to update the
  *     value displayed to the user.
@@ -162,29 +194,44 @@ export const update = (
   sources: interfaces.Sources,
   figures: interfaces.Figures,
   tooltips: interfaces.Tooltips,
+  widgets: interfaces.Widgets,
 ): number => {
-  const computedData = computeData(data, bwFactor, hdiProbability);
-  for (let i = 0; i < figureNames.length; i += 1) {
-    // Update all sources with new data calculated above.
-    const figureName = figureNames[i];
-    sources[figureName].distribution.data = {
-      x: computedData[figureName].distribution.x,
-      y: computedData[figureName].distribution.y,
-    };
-    sources[figureName].hdi.data = {
-      base: computedData[figureName].hdi.base,
-      lower: computedData[figureName].hdi.lower,
-      upper: computedData[figureName].hdi.upper,
-    };
-    sources[figureName].stats.data = computedData[figureName].stats;
-    sources[figureName].labels.data = computedData[figureName].labels;
+  const activeStatistic = widgets.stats_button.active as number;
+  const computedData = computeData(data, bwFactor, hdiProbability, activeStatistic);
 
-    // Update the axes labels.
-    updateAxisLabel(figures[figureName].below[0], rvName);
+  // Marginal figure.
+  // eslint-disable-next-line prefer-destructuring
+  const bandwidth = computedData.marginal.distribution.bandwidth;
+  sources.marginal.distribution.data = {
+    x: computedData.marginal.distribution.x,
+    y: computedData.marginal.distribution.y,
+  };
+  sources.marginal.hdi.data = {
+    base: computedData.marginal.hdi.base,
+    lower: computedData.marginal.hdi.lower,
+    upper: computedData.marginal.hdi.upper,
+  };
+  sources.marginal.stats.data = computedData.marginal.stats;
+  sources.marginal.labels.data = computedData.marginal.labels;
+  tooltips.marginal.distribution.tooltips = [[rvName, '@x']];
+  tooltips.marginal.stats.tooltips = [['', '@text']];
+  updateAxisLabel(figures.marginal.below[0] as Axis, rvName);
 
-    // Update the tooltips.
-    tooltips[figureName].stats.tooltips = [['', '@text']];
-    tooltips[figureName].distribution.tooltips = [[rvName, '@x']];
-  }
-  return computedData.marginal.distribution.bandwidth;
+  // Cumulative figure.
+  sources.cumulative.distribution.data = {
+    x: computedData.cumulative.distribution.x,
+    y: computedData.cumulative.distribution.y,
+  };
+  sources.cumulative.hdi.data = {
+    base: computedData.cumulative.hdi.base,
+    lower: computedData.cumulative.hdi.lower,
+    upper: computedData.cumulative.hdi.upper,
+  };
+  sources.cumulative.stats.data = computedData.cumulative.stats;
+  sources.cumulative.labels.data = computedData.cumulative.labels;
+  tooltips.cumulative.distribution.tooltips = [[rvName, '@x']];
+  tooltips.cumulative.stats.tooltips = [['', '@text']];
+  updateAxisLabel(figures.cumulative.below[0] as Axis, rvName);
+
+  return bandwidth;
 };

src/beanmachine/ppl/diagnostics/tools/js/src/marginal1d/interfaces.ts

@@ -8,6 +8,10 @@
 import {Plot} from '@bokehjs/models/plots/plot';
 import {ColumnDataSource} from '@bokehjs/models/sources/column_data_source';
 import {HoverTool} from '@bokehjs/models/tools/inspectors/hover_tool';
+import {Div} from '@bokehjs/models/widgets/div';
+import {RadioButtonGroup} from '@bokehjs/models/widgets/radio_button_group';
+import {Select} from '@bokehjs/models/widgets/selectbox';
+import {Slider} from '@bokehjs/models/widgets/slider';
 
 // NOTE: In the corresponding Python typing files for the diagnostic tool, we define
 //       similar types using a TypedDict object. TypeScript allows us to maintain
@@ -95,3 +99,11 @@ export interface Tooltips {
   marginal: Tooltip;
   cumulative: Tooltip;
 }
+
+export interface Widgets {
+  rv_select: Select;
+  bw_factor_slider: Slider;
+  bw_div: Div;
+  hdi_slider: Slider;
+  stats_button: RadioButtonGroup;
+}

src/beanmachine/ppl/diagnostics/tools/js/src/stats/array.ts

@@ -5,6 +5,29 @@
  * LICENSE file in the root directory of this source tree.
  */
 
+/**
+ * Syntactic sugar for summing an array of numbers.
+ *
+ * @param {number[]} data - The array of data.
+ * @returns {number} The sum of the array of data.
+ */
+export const arraySum = (data: number[]): number => {
+  return data.reduce((previousValue, currentValue) => {
+    return previousValue + currentValue;
+  });
+};
+
+/**
+ * Calculate the mean of the given array of data.
+ *
+ * @param {number[]} data - The array of data.
+ * @returns {number} The mean of the given data.
+ */
+export const arrayMean = (data: number[]): number => {
+  const dataSum = arraySum(data);
+  return dataSum / data.length;
+};
+
 /**
  * Cumulative sum of the given data.
  *
@@ -128,3 +151,23 @@ export const valueCounts = (data: number[]): {[key: string]: number} => {
   }
   return counts;
 };
+
+/**
+ * Calculate the median value for the given array.
+ *
+ * @param {number[]} data - Numerical array of data.
+ * @returns {number} The median value of the given data.
+ */
+export const arrayMedian = (data: number[]): number => {
+  const sortedArray = numericalSort(data);
+  const arrayLength = sortedArray.length;
+  const isEven = sortedArray.length % 2 === 0;
+  let median;
+  if (isEven) {
+    const index = arrayLength / 2;
+    median = (sortedArray[index - 1] + sortedArray[index]) / 2;
+  } else {
+    median = sortedArray[Math.floor(arrayLength / 2)];
+  }
+  return median;
+};

src/beanmachine/ppl/diagnostics/tools/marginal1d/tool.py

@@ -24,14 +24,14 @@ class Marginal1d(DiagnosticToolBaseClass):
 
     Attributes:
         data (Dict[str, List[List[float]]]): JSON serializable representation of the
-            given `mcs` object.
+            given ``mcs`` object.
         rv_names (List[str]): The list of random variables string names for the given
             model.
         num_chains (int): The number of chains of the model.
         num_draws (int): The number of draws of the model for each chain.
         palette (List[str]): A list of color values used for the glyphs in the figures.
-            The colors are specifically chosen from the Colorblind palette defined in
-            Bokeh.
+            The colors are specifically chosen from the ``Colorblind`` palette defined
+            in Bokeh.
         tool_js (str):The JavaScript callbacks needed to render the Bokeh tool
             independently from a Python server.
     """
@@ -40,6 +40,12 @@ def __init__(self: Marginal1d, mcs: MonteCarloSamples) -> None:
         super(Marginal1d, self).__init__(mcs)
 
     def create_document(self: Marginal1d) -> Model:
+        """
+        Create the Bokeh document for the diagnostic tool.
+
+        Returns:
+            Model: A Bokeh Model object.
+        """
         # Initialize widget values using Python.
         rv_name = self.rv_names[0]
         bw_factor = 1.0
@@ -110,6 +116,7 @@ def create_document(self: Marginal1d) -> Model:
                 sources,
                 figures,
                 tooltips,
+                widgets,
               );
             }} catch (error) {{
               {self.tool_js}
@@ -121,6 +128,7 @@ def create_document(self: Marginal1d) -> Model:
                 sources,
                 figures,
                 tooltips,
+                widgets,
               );
             }}
         """
@@ -135,6 +143,7 @@ def create_document(self: Marginal1d) -> Model:
             "figures": figures,
             "tooltips": tooltips,
             "toolView": tool_view,
+            "widgets": widgets,
         }
 
         # Each widget requires slightly different JS, except for the sliders.
@@ -155,10 +164,12 @@ def create_document(self: Marginal1d) -> Model:
         """
         rv_select_callback = CustomJS(args=callback_arguments, code=rv_select_js)
         slider_callback = CustomJS(args=callback_arguments, code=slider_js)
+        button_callback = CustomJS(args=callback_arguments, code=slider_js)
 
         # Tell Python to use the JavaScript.
         widgets["rv_select"].js_on_change("value", rv_select_callback)
         widgets["bw_factor_slider"].js_on_change("value", slider_callback)
         widgets["hdi_slider"].js_on_change("value", slider_callback)
+        widgets["stats_button"].js_on_change("active", button_callback)
 
         return tool_view