feat(ava/lite-insight): 新增多维度下钻波动归因方法

packages/ava/__tests__/unit/insight/extractors/causalInference/drillDown.test.ts

@@ -0,0 +1,75 @@
+import { dimensionDrillDownAttribution } from '../../../../../src/insight/insights/extractors/causalInference/dimensionDrillDown';
+
+import type { DrillDownProps } from '../../../../../src/insight/insights/extractors/causalInference/types';
+
+const data = [
+  {
+    City: 'huangshishi',
+    Province: 'shanxi',
+    ClientGender: 'male',
+    OrderDate: '2022/2/19',
+    Price: 13722.76,
+    UnitCost: 292.72,
+  },
+  {
+    City: 'huangshishi',
+    Province: 'shanxi',
+    ClientGender: 'male',
+    OrderDate: '2022/2/21',
+    Price: 24020.88,
+    UnitCost: 5447.57,
+  },
+  {
+    City: 'huangshishi',
+    Province: 'shanxi',
+    ClientGender: 'male',
+    OrderDate: '2022/2/16',
+    Price: 40145.8,
+    UnitCost: 3696.42,
+  },
+  {
+    City: 'huanggangshi',
+    Province: 'qinghai',
+    ClientGender: 'male',
+    OrderDate: '2022/2/18',
+    Price: 99980.16,
+    UnitCost: 34393.38,
+  },
+  {
+    City: 'huanggangshi',
+    Province: 'qinghai',
+    ClientGender: 'ma le',
+    OrderDate: '2022/2/17',
+    Price: 12656.11,
+    UnitCost: 6012.96,
+  },
+  {
+    City: 'huanggangshi',
+    Province: 'qinghai',
+    ClientGender: 'male',
+    OrderDate: '2022/2/19',
+    Price: 42464.05,
+    UnitCost: 3113.68,
+  },
+];
+describe('MultiDim Test', () => {
+  test('check the disassemable result', () => {
+    const props: DrillDownProps = {
+      sourceData: data,
+      dimensions: ['Province', 'City'],
+      targetMeasure: 'Price',
+      timeSeriesDim: 'OrderDate',
+      baseInterval: {
+        startPoint: '2022/2/17',
+        endPoint: '2022/2/18',
+      },
+      currInterval: {
+        startPoint: '2022/2/19',
+        endPoint: '2022/2/20',
+      },
+    };
+    const tempResult = dimensionDrillDownAttribution(props);
+    const treeResult = tempResult.resultInTree;
+    expect(treeResult?.City.huangshishi.info.currValue).toBe(13722.76);
+  });
+});

packages/ava/package.json

@@ -60,7 +60,9 @@
     "heap-js": "^2.1.6",
     "lodash": "^4.17.21",
     "regression": "^2.0.1",
-    "tslib": "^2.3.1"
+    "tslib": "^2.3.1",
+    "expr-eval": "^2.0.2",
+    "moment": "^2.29.4"
   },
   "devDependencies": {
     "@antv/algorithm": "^0.1.25",

packages/ava/src/index.ts

@@ -93,7 +93,7 @@ export type {
 } from './data';
 
 /* insight */
-export { getInsights, generateInsightVisualizationSpec } from './insight';
+export { getInsights, generateInsightVisualizationSpec, dimensionDrillDownAttribution } from './insight';
 export type {
   Datum,
   DomainType,
@@ -124,6 +124,7 @@ export type {
   LowVarianceInfo,
   CorrelationInfo,
   InsightsResult,
+  DimensionDrillDownProps,
 } from './insight';
 
 /* NTV (Narrative Text Vis) */

packages/ava/src/insight/index.ts

@@ -1,3 +1,5 @@
 export { getInsights } from './pipeline';
 export { generateInsightVisualizationSpec } from './pipeline/visualize';
+export { dimensionDrillDownAttribution } from './insights/extractors/causalInference/dimensionDrillDown';
+export type { DrillDownProps as DimensionDrillDownProps } from './insights/extractors/causalInference/types';
 export * from './types';

packages/ava/src/insight/insights/extractors/causalInference/dimensionDrillDown.ts

@@ -0,0 +1,54 @@
+import { locatedInInterval, enumerateAllDimensionCombinationsByDFS } from './util';
+
+import type { InfoType, DimensionDrillDownResult, TreeDim, DataLocation, DrillDownProps } from './types';
+
+/** Main function for dimension drill down attribution */
+export const dimensionDrillDownAttribution = ({
+  sourceData,
+  dimensions,
+  targetMeasure: measure,
+  timeSeriesDim: fluctuationDim,
+  baseInterval,
+  currInterval,
+}: DrillDownProps): DimensionDrillDownResult => {
+  /** remove invalid data */
+  const data = sourceData.filter((item) => !Object.values(item).some((v) => v === null || v === undefined));
+
+  const globalDiff: InfoType = {
+    baseValue: 0,
+    currValue: 0,
+    diff: 0,
+  };
+
+  const resultTree: TreeDim = {};
+  const DictFlatten = {};
+  /** traverse the input data and build the result data structure; */
+  data.forEach((item) => {
+    let location: DataLocation = 'none';
+    if (locatedInInterval(item[fluctuationDim], baseInterval.startPoint, baseInterval.endPoint)) {
+      location = 'left';
+      globalDiff.baseValue += item[measure] as number;
+    }
+    if (locatedInInterval(item[fluctuationDim], currInterval.startPoint, currInterval.endPoint)) {
+      location = 'right';
+      globalDiff.currValue += item[measure] as number;
+    }
+    if (location !== 'none') {
+      const deque: string[] = [];
+      enumerateAllDimensionCombinationsByDFS(
+        item,
+        0,
+        dimensions,
+        resultTree,
+        DictFlatten,
+        deque,
+        measure,
+        fluctuationDim,
+        location
+      );
+    }
+  });
+  globalDiff.diff = globalDiff.currValue - globalDiff.baseValue;
+
+  return { resultInTree: resultTree, globalDiff, resultInList: Object.values(DictFlatten) };
+};

packages/ava/src/insight/insights/extractors/causalInference/types.ts

@@ -0,0 +1,133 @@
+import type { Datum } from '../../../types';
+
+/** DataConfig specifies the input data with its focused dimensions and target measure to be analysed. */
+export type DataConfig = {
+  sourceData: Datum[];
+  dimensions: string[];
+  measures: string[];
+  expression?: string;
+};
+
+/** FluctInfo is the necessary input for fluctuation analysis.  */
+export type FluctInfo = {
+  /** Typically, flucDim is a time dimension that the measure value varies on */
+  fluctDim: string;
+  /** baseInterval is the time interval that is assigned as the basement */
+  baseInterval: CompareInterval;
+  /** currInterval is the time interval that user is focusing on */
+  currInterval: CompareInterval;
+};
+
+/** Time Interval in Fluctuation Analysis */
+export type CompareInterval = {
+  /** start time of this interval */
+  startPoint: string | number;
+  /** end time of this interval */
+  endPoint: string | number;
+};
+
+/** a flag that indicates the belongings of a single line of data, which is useful for aggregation */
+export type DataLocation = 'left' | 'right' | 'none';
+
+/** Record the calculation result */
+export type InfoType = {
+  /** Calculation value corresponding to baseInterval */
+  baseValue: number;
+  /** Calculation value corresponding to currInterval */
+  currValue: number;
+  /** diff = currValue - baseValue */
+  diff: number;
+};
+
+/**  Dimension drill down Result type */
+export type DimensionDrillDownResult = {
+  /** total difference  */
+  globalDiff: InfoType;
+  /** Tree like returned data */
+  resultInTree?: TreeDim;
+  /** Flatten returned data */
+  resultInList?: FlattenResult[];
+};
+
+/**  Dimension drill down Result type */
+export type MeasureDecomposeResult = {
+  /** total difference  */
+  globalDiff: InfoType;
+
+  resultByMeasure?: FunctionBasedResult;
+};
+
+/** The first string is dimension name, the second string is dimension value */
+export type DimWithValue = Record<string, string>;
+
+/** Dimension drill down attribution Result type that has been formalized into Datum[] */
+export type FlattenResult = Partial<DimWithValue & InfoType>;
+
+/** Function based attribution Result type */
+export type FunctionBasedResult = Record<string, InfoType>;
+
+/** Example: Tree Data Structure for dimension drill down analysis
+ *
+ * {
+ * dimNameA: {
+ *   dimValueA1: {
+ *      info: {baseValue: 222, curValue: 555, diff: 333},
+ *      drillDown: {
+ *          dimNameB: {
+ *   dimValueB1: {
+ *      info: {baseValue: 000, curValue: 111, diff: 111},
+ *      drillDown: {}
+ *   },
+ *   dimValueB2: {
+ *      info: {baseValue: 222, curValue: 444, diff: 222},
+ *      drillDown: {}
+ *   }
+ * }
+ *      }
+ *   },
+ * },
+ * dimNameB: {
+ *   dimValueB1: {
+ *      info: {baseValue: 000, curValue: 111, diff: 111},
+ *      drillDown: {}
+ *   },
+ *   dimValueB2: {
+ *      info: {baseValue: 222, curValue: 444, diff: 222},
+ *      drillDown: {}
+ *   }
+ * }
+ * }
+ *
+ */
+
+/** The dimension name level of the tree or the first level in drillDown */
+export interface TreeDim {
+  [dimName: string]: TreeDimVal;
+}
+
+/** The dimension value level of the tree which is inside the dimension name level */
+export type TreeDimVal = Record<string | number, TreeDrillDown>;
+
+/** The information level inside the dimension value  */
+export interface TreeDrillDown {
+  info: InfoType;
+  drillDown: TreeDim;
+}
+
+interface CausalInferenceProps {
+  sourceData: Datum[];
+  dimensions: string[];
+  timeSeriesDim: string;
+  baseInterval: CompareInterval;
+  currInterval: CompareInterval;
+}
+
+/** Dimension drill down based attribution function props  */
+export interface DrillDownProps extends CausalInferenceProps {
+  targetMeasure: string;
+}
+
+export interface MeasureDecomposeProps extends CausalInferenceProps {
+  allMeasures: string[];
+  expression: string;
+}