Implement stretching

src/lib/data/analyze.ts

@@ -0,0 +1,92 @@
+import {
+    type ActivityStandards,
+    type Levels,
+} from "$lib/services/calculator/main";
+import { Activity, clamp, Gender, range } from "$lib/services/calculator/util";
+import allStandards from "./standards.json";
+import { levenbergMarquardt as LM } from "ml-levenberg-marquardt";
+
+const activityStandards = allStandards as ActivityStandards;
+
+for (const activity of Object.values(Activity)) {
+    const DATASET_MAX_LEVEL = 5;
+    const NEW_LEVEL_COUNT = 2;
+    const RATIO_CLAMP = 1.4;
+
+    function expDecayModel([A, B, C]: number[]) {
+        return (i: number) => A * Math.exp(-B * i) + C;
+    }
+
+    for (const gender of Object.values(Gender)) {
+        const standardsByGender = activityStandards[activity].standards.filter(
+            (s) => s["metrics"].gender === gender,
+        );
+        const ages = [...new Set(standardsByGender.map((s) => s.metrics.age))];
+        for (const age of ages) {
+            const standards = standardsByGender.filter(
+                (s) => s.metrics.age === age,
+            );
+
+            const data = {
+                x: [] as number[],
+                y: [] as number[],
+            };
+
+            const isIncreasing =
+                standards[0].levels["2"] > standards[0].levels["1"];
+
+            for (const i of range(DATASET_MAX_LEVEL).slice(0, -1)) {
+                const level = i + 1;
+                const progressionRatios: number[] = [];
+                for (const standard of standards) {
+                    const curr = standard.levels[level];
+                    const next = standard.levels[level];
+                    progressionRatios.push(
+                        isIncreasing ? next / curr : curr / next,
+                    );
+                }
+
+                const sum = progressionRatios.reduce((p, c) => p + c, 0);
+                const progressionRatioAvg = sum / progressionRatios.length;
+
+                data.x.push(i);
+                data.y.push(progressionRatioAvg);
+            }
+
+            const fittedParams = LM(data, expDecayModel, {
+                initialValues: [0.4, 0.5, 1.1],
+                minValues: [0.0, 0.0, 1.02],
+                maxValues: [1.0, 2.0, 1.25],
+                maxIterations: 200,
+            });
+            const getDecayRatio = (i: number) =>
+                clamp(
+                    expDecayModel(fittedParams.parameterValues)(i),
+                    1,
+                    RATIO_CLAMP,
+                );
+
+            // UPDATE THE DATA
+            for (const standard of standards) {
+                const newLevels: Levels = {};
+
+                let prev = standard.levels["1"];
+                for (const i of range(NEW_LEVEL_COUNT)) {
+                    const level = i + 1;
+                    const ratio = getDecayRatio(level - NEW_LEVEL_COUNT);
+                    prev = isIncreasing ? prev / ratio : prev * ratio;
+                    newLevels[level] = Math.round(prev);
+                }
+
+                const oldLevels = standard.levels as Levels;
+                for (const i of range(DATASET_MAX_LEVEL)) {
+                    const level = i + 1;
+                    newLevels[level + NEW_LEVEL_COUNT] = oldLevels[level];
+                }
+                standard.levels = newLevels;
+                console.log(newLevels);
+            }
+        }
+    }
+}
+// Bun.write("./newStandards.json", JSON.stringify(allStandards, null, 2));