Update ByteEfficiencyAudits to use log-normal distribution scoring (#…

…11883)
GoogleChrome · May 20, 2023 · f76e799 · f76e799
1 parent 9922e4c
commit f76e799
Show file tree

Hide file tree

Showing 2 changed files with 22 additions and 26 deletions.
diff --git a/core/audits/byte-efficiency/byte-efficiency-audit.js b/core/audits/byte-efficiency/byte-efficiency-audit.js
@@ -5,7 +5,6 @@
  */
 
 import {Audit} from '../audit.js';
-import {linearInterpolation} from '../../lib/statistics.js';
 import {LanternInteractive} from '../../computed/metrics/lantern-interactive.js';
 import * as i18n from '../../lib/i18n/i18n.js';
 import {NetworkRecords} from '../../computed/network-records.js';
@@ -17,9 +16,13 @@ const str_ = i18n.createIcuMessageFn(import.meta.url, {});
 /** @typedef {import('../../lib/dependency-graph/simulator/simulator').Simulator} Simulator */
 /** @typedef {import('../../lib/dependency-graph/base-node.js').Node} Node */
 
-const WASTED_MS_FOR_AVERAGE = 300;
-const WASTED_MS_FOR_POOR = 750;
-const WASTED_MS_FOR_SCORE_OF_ZERO = 5000;
+// Parameters for log-normal distribution scoring. These values were determined by fitting the
+// log-normal cumulative distribution function curve to the former method of linear interpolation
+// scoring between the control points {average = 300 ms, poor = 750 ms, zero = 5000 ms} using the
+// curve-fit tool at https://mycurvefit.com/ rounded to the nearest integer. See
+// https://www.desmos.com/calculator/gcexiyesdi for an interactive visualization of the curve fit.
+const WASTED_MS_P10 = 150;
+const WASTED_MS_MEDIAN = 935;
 
 /**
  * @typedef {object} ByteEfficiencyProduct
@@ -38,26 +41,18 @@ const WASTED_MS_FOR_SCORE_OF_ZERO = 5000;
  */
 class ByteEfficiencyAudit extends Audit {
   /**
-   * Creates a score based on the wastedMs value using linear interpolation between control points.
-   * A negative wastedMs is scored as 1, assuming time is not being wasted with respect to the
-   * opportunity being measured.
+   * Creates a score based on the wastedMs value using log-normal distribution scoring. A negative
+   * wastedMs will be scored as 1, assuming time is not being wasted with respect to the opportunity
+   * being measured.
    *
    * @param {number} wastedMs
    * @return {number}
    */
   static scoreForWastedMs(wastedMs) {
-    if (wastedMs <= 0) {
-      return 1;
-    } else if (wastedMs < WASTED_MS_FOR_AVERAGE) {
-      return linearInterpolation(0, 1, WASTED_MS_FOR_AVERAGE, 0.75, wastedMs);
-    } else if (wastedMs < WASTED_MS_FOR_POOR) {
-      return linearInterpolation(WASTED_MS_FOR_AVERAGE, 0.75, WASTED_MS_FOR_POOR, 0.5, wastedMs);
-    } else {
-      return Math.max(
-        0,
-        linearInterpolation(WASTED_MS_FOR_POOR, 0.5, WASTED_MS_FOR_SCORE_OF_ZERO, 0, wastedMs)
-      );
-    }
+    return Audit.computeLogNormalScore(
+      {p10: WASTED_MS_P10, median: WASTED_MS_MEDIAN},
+      wastedMs
+    );
   }
 
   /**

diff --git a/core/test/audits/byte-efficiency/byte-efficiency-audit-test.js b/core/test/audits/byte-efficiency/byte-efficiency-audit-test.js
@@ -414,13 +414,14 @@ describe('Byte efficiency base audit', () => {
     it('scores wastedMs values', () => {
       expect(ByteEfficiencyAudit.scoreForWastedMs(-50)).toBe(1);
       expect(ByteEfficiencyAudit.scoreForWastedMs(0)).toBe(1);
-      expect(ByteEfficiencyAudit.scoreForWastedMs(240)).toBe(0.8);
-      expect(ByteEfficiencyAudit.scoreForWastedMs(300)).toBe(0.75);
-      expect(ByteEfficiencyAudit.scoreForWastedMs(390)).toBe(0.7);
-      expect(ByteEfficiencyAudit.scoreForWastedMs(750)).toBe(0.5);
-      expect(ByteEfficiencyAudit.scoreForWastedMs(1_175)).toBe(0.45);
-      expect(ByteEfficiencyAudit.scoreForWastedMs(5_000)).toBe(0);
-      expect(ByteEfficiencyAudit.scoreForWastedMs(10_000)).toBe(0);
+      expect(ByteEfficiencyAudit.scoreForWastedMs(240)).toBe(0.82);
+      expect(ByteEfficiencyAudit.scoreForWastedMs(300)).toBe(0.78);
+      expect(ByteEfficiencyAudit.scoreForWastedMs(390)).toBe(0.72);
+      expect(ByteEfficiencyAudit.scoreForWastedMs(750)).toBe(0.56);
+      expect(ByteEfficiencyAudit.scoreForWastedMs(1_175)).toBe(0.43);
+      expect(ByteEfficiencyAudit.scoreForWastedMs(5_000)).toBe(0.12);
+      expect(ByteEfficiencyAudit.scoreForWastedMs(10_000)).toBe(0.04);
+      expect(ByteEfficiencyAudit.scoreForWastedMs(30_000)).toBe(0);
       expect(ByteEfficiencyAudit.scoreForWastedMs(Number.MAX_VALUE)).toBe(0);
     });
   });