ggaction 0.0.17

Published-release documentation · 281 actions; release compatibility · contract 5520b0fa9525

Repair missing observations

Restore an omitted time key, impute its measure, and recompute a trailing window when new source observations arrive.

Repair missing observations shown as an executable chart workflow
Repair missing observations. Each panel is generated from the complete program below.

Prerequisites

Use the Getting Started browser module setup, install the full ggaction entry, and provide <canvas id="chart"></canvas>. This complete example includes its data and imports. These APIs require ggaction 0.0.15 or later; check the documentation version for release compatibility.

Decision sequence: line chart → data transform actions → mark rebind.

Complete program

import { chart, hconcat } from "ggaction";
import { render } from "ggaction";

export function createMissingObservationsWorkflow() {
  const rows = [{ t: 1, value: 2 }, { t: 3, value: 6 }];
  function derive(program, source, suffix) {
    return program
      .createCompleteData({ id: `complete${suffix}`, source, key: "t", values: [1, 2, 3] })
      .createImputedData({ id: `imputed${suffix}`, source: `complete${suffix}`,
        fields: "value", sortBy: [{ field: "t" }], method: "linear" })
      .createWindowData({ id: `window${suffix}`, source: `imputed${suffix}`,
        sortBy: [{ field: "t" }], operations: [
          { op: "movingMean", field: "value", as: "mean", frame: { preceding: 1 } }
        ] });
  }
  const before = derive(chart()
    .createCanvas({ width: 420, height: 300, margin: 65 })
    .createData({ id: "raw", values: rows }), "raw", "Before")
    .createLinePlot({ id: "trend", data: "windowBefore", x: "t", y: "mean" });
  const after = derive(before.createData({ id: "replacement", values: [
    { t: 1, value: 4 }, { t: 3, value: 8 }
  ] }), "replacement", "After")
    .bindMarkData({ target: "trend", data: "windowAfter" });
  return hconcat({ programs: [
    before.createTitle({ text: "Before: moving means 2, 3, 5" }),
    after.createTitle({ text: "After: moving means 4, 5, 7" })
  ], gap: 20 });
}

const program = createMissingObservationsWorkflow();
render(program, document.querySelector("#chart").getContext("2d"));

Expected result

The source has two rows: (t=1,value=2) and (t=3,value=6). Complete produces three rows with a null at t=2; linear imputation produces [2,4,6]; a preceding-one-row mean produces [2,3,5]. Replacement values [4,8] produce [4,6,8], then [4,5,7]. The first program and its original rows stay unchanged.

Boundaries and failure behavior

Completion adds keys; imputation fills cells and does not add rows. Each observed group/key must be unique. Linear imputation uses position distance, not index distance. Source rows are immutable: there is no general edit-source-data action. Create the replacement, rebuild its derivation, then bind an eligible independent mark. To change a transform decision on the same source, use its focused editor with dependents: "recompute".

Authoring conventions · Data policies · Complete action reference