[FIX] charts: non-invertible matrix returns NaN as predicted data

Task Description When trying to make a trending line, almost all model use matrix computation to get the predicted data, with some matrix multiplications and inversions. When the matrixes aren't invertible the user get a traceback, which is not fine. This PR aims to fix these issues by returning an array of NaN when there is an error in the computation of the predicted dataset. Related Task closes #5202 Task: 4328743 Signed-off-by: Lucas Lefèvre (lul) <[email protected]>
odoo · Dec 2, 2024 · 02b6618 · 02b6618
1 parent 6e043e4
commit 02b6618
Show file tree

Hide file tree

Showing 3 changed files with 52 additions and 31 deletions.
diff --git a/src/components/side_panel/chart/chart_with_axis/design_panel.ts b/src/components/side_panel/chart/chart_with_axis/design_panel.ts
@@ -195,7 +195,7 @@ export class ChartWithAxisDesignPanel<P extends Props = Props> extends Component
       case "polynomial":
         config = {
           type: "polynomial",
-          order: type === "linear" ? 1 : 2,
+          order: type === "linear" ? 1 : this.getMaxPolynomialDegree(),
         };
         break;
       case "exponential":

diff --git a/src/helpers/figures/charts/chart_common.ts b/src/helpers/figures/charts/chart_common.ts
@@ -535,41 +535,48 @@ export function interpolateData(
   const labelRange = labelMax - labelMin;
   const normalizedLabels = labels.map((v) => (v - labelMin) / labelRange);
   const normalizedNewLabels = newLabels.map((v) => (v - labelMin) / labelRange);
-  switch (config.type) {
-    case "polynomial": {
-      const order = config.order ?? 2;
-      if (order === 1) {
-        return predictLinearValues([values], [normalizedLabels], [normalizedNewLabels], true)[0];
+  try {
+    switch (config.type) {
+      case "polynomial": {
+        const order = config.order;
+        if (!order) {
+          return Array.from({ length: newLabels.length }, () => NaN);
+        }
+        if (order === 1) {
+          return predictLinearValues([values], [normalizedLabels], [normalizedNewLabels], true)[0];
+        }
+        const coeffs = polynomialRegression(values, normalizedLabels, order, true).flat();
+        return normalizedNewLabels.map((v) => evaluatePolynomial(coeffs, v, order));
       }
-      const coeffs = polynomialRegression(values, normalizedLabels, order, true).flat();
-      return normalizedNewLabels.map((v) => evaluatePolynomial(coeffs, v, order));
-    }
-    case "exponential": {
-      const positiveLogValues: number[] = [];
-      const filteredLabels: number[] = [];
-      for (let i = 0; i < values.length; i++) {
-        if (values[i] > 0) {
-          positiveLogValues.push(Math.log(values[i]));
-          filteredLabels.push(normalizedLabels[i]);
+      case "exponential": {
+        const positiveLogValues: number[] = [];
+        const filteredLabels: number[] = [];
+        for (let i = 0; i < values.length; i++) {
+          if (values[i] > 0) {
+            positiveLogValues.push(Math.log(values[i]));
+            filteredLabels.push(normalizedLabels[i]);
+          }
+        }
+        if (!filteredLabels.length) {
+          return Array.from({ length: newLabels.length }, () => NaN);
         }
+        return expM(
+          predictLinearValues([positiveLogValues], [filteredLabels], [normalizedNewLabels], true)
+        )[0];
       }
-      if (!filteredLabels.length) {
-        return [];
+      case "logarithmic": {
+        return predictLinearValues(
+          [values],
+          logM([normalizedLabels]),
+          logM([normalizedNewLabels]),
+          true
+        )[0];
       }
-      return expM(
-        predictLinearValues([positiveLogValues], [filteredLabels], [normalizedNewLabels], true)
-      )[0];
-    }
-    case "logarithmic": {
-      return predictLinearValues(
-        [values],
-        logM([normalizedLabels]),
-        logM([normalizedNewLabels]),
-        true
-      )[0];
+      default:
+        return [];
     }
-    default:
-      return [];
+  } catch (e) {
+    return Array.from({ length: newLabels.length }, () => NaN);
   }
 }
 

diff --git a/tests/figures/chart/chart_plugin.test.ts b/tests/figures/chart/chart_plugin.test.ts
@@ -3209,6 +3209,20 @@ describe("trending line", () => {
     }
   });
 
+  test("non-invertible matrix doesn't throw error", () => {
+    // prettier-ignore
+    setGrid(model, {
+      A1: "label",
+      A2: "0",
+      A3: "1",
+    });
+    updateChart(model, "1", {
+      dataSets: [{ dataRange: "A1:A3", trend: { display: true, type: "polynomial", order: 2 } }],
+    });
+    const runtime = model.getters.getChartRuntime("1") as LineChartRuntime;
+    expect(runtime.chartJsConfig.data.datasets[1].data.every((x) => isNaN(Number(x)))).toBeTruthy();
+  });
+
   test("trend line works with real date values as labels", () => {
     setGrid(model, {
       B1: "1",