[ML] DF Analytics: create classification jobs results view (elastic#5…

…2584)

* wip: create classification results page + table and evaluate panel

* enable view link for classification jobs

* wip: fetch classification eval data

* wip: display confusion matrix in datagrid

* evaluate panel: add heatmap for cells and doc count

* Update use of loadEvalData in expanded row component

* Add metric type for evaluate endpoint and fix localization error

* handle no incorrect prediction classes case for confusion matrix. remove unused translation

* setCellProps needs to be called from a lifecycle method - wrap in useEffect

* TypeScript improvements

* fix datagrid column resize affecting results table

* allow custom prediction field for classification jobs

* ensure values are rounded correctly and add tooltip

* temp workaroun for datagrid width issues

x-pack/legacy/plugins/ml/public/application/data_frame_analytics/_index.scss

@@ -1,5 +1,6 @@
 @import 'pages/analytics_exploration/components/exploration/index';
 @import 'pages/analytics_exploration/components/regression_exploration/index';
+@import 'pages/analytics_exploration/components/classification_exploration/index';
 @import 'pages/analytics_management/components/analytics_list/index';
 @import 'pages/analytics_management/components/create_analytics_form/index';
 @import 'pages/analytics_management/components/create_analytics_flyout/index';

x-pack/legacy/plugins/ml/public/application/data_frame_analytics/common/analytics.ts

@@ -35,6 +35,7 @@ interface ClassificationAnalysis {
     dependent_variable: string;
     training_percent?: number;
     num_top_classes?: string;
+    prediction_field_name?: string;
   };
 }
 
@@ -74,13 +75,33 @@ export interface RegressionEvaluateResponse {
   };
 }
 
+export interface PredictedClass {
+  predicted_class: string;
+  count: number;
+}
+
+export interface ConfusionMatrix {
+  actual_class: string;
+  actual_class_doc_count: number;
+  predicted_classes: PredictedClass[];
+  other_predicted_class_doc_count: number;
+}
+
+export interface ClassificationEvaluateResponse {
+  classification: {
+    multiclass_confusion_matrix: {
+      confusion_matrix: ConfusionMatrix[];
+    };
+  };
+}
+
 interface GenericAnalysis {
   [key: string]: Record<string, any>;
 }
 
 interface LoadEvaluateResult {
   success: boolean;
-  eval: RegressionEvaluateResponse | null;
+  eval: RegressionEvaluateResponse | ClassificationEvaluateResponse | null;
   error: string | null;
 }
 
@@ -109,6 +130,7 @@ export const getAnalysisType = (analysis: AnalysisConfig) => {
 
 export const getDependentVar = (analysis: AnalysisConfig) => {
   let depVar = '';
+
   if (isRegressionAnalysis(analysis)) {
     depVar = analysis.regression.dependent_variable;
   }
@@ -124,17 +146,26 @@ export const getPredictionFieldName = (analysis: AnalysisConfig) => {
   let predictionFieldName;
   if (isRegressionAnalysis(analysis) && analysis.regression.prediction_field_name !== undefined) {
     predictionFieldName = analysis.regression.prediction_field_name;
+  } else if (
+    isClassificationAnalysis(analysis) &&
+    analysis.classification.prediction_field_name !== undefined
+  ) {
+    predictionFieldName = analysis.classification.prediction_field_name;
   }
   return predictionFieldName;
 };
 
-export const getPredictedFieldName = (resultsField: string, analysis: AnalysisConfig) => {
+export const getPredictedFieldName = (
+  resultsField: string,
+  analysis: AnalysisConfig,
+  forSort?: boolean
+) => {
   // default is 'ml'
   const predictionFieldName = getPredictionFieldName(analysis);
   const defaultPredictionField = `${getDependentVar(analysis)}_prediction`;
   const predictedField = `${resultsField}.${
     predictionFieldName ? predictionFieldName : defaultPredictionField
-  }`;
+  }${isClassificationAnalysis(analysis) && !forSort ? '.keyword' : ''}`;
   return predictedField;
 };
 
@@ -153,13 +184,32 @@ export const isClassificationAnalysis = (arg: any): arg is ClassificationAnalysi
   return keys.length === 1 && keys[0] === ANALYSIS_CONFIG_TYPE.CLASSIFICATION;
 };
 
-export const isRegressionResultsSearchBoolQuery = (
-  arg: any
-): arg is RegressionResultsSearchBoolQuery => {
+export const isResultsSearchBoolQuery = (arg: any): arg is ResultsSearchBoolQuery => {
   const keys = Object.keys(arg);
   return keys.length === 1 && keys[0] === 'bool';
 };
 
+export const isRegressionEvaluateResponse = (arg: any): arg is RegressionEvaluateResponse => {
+  const keys = Object.keys(arg);
+  return (
+    keys.length === 1 &&
+    keys[0] === ANALYSIS_CONFIG_TYPE.REGRESSION &&
+    arg?.regression?.mean_squared_error !== undefined &&
+    arg?.regression?.r_squared !== undefined
+  );
+};
+
+export const isClassificationEvaluateResponse = (
+  arg: any
+): arg is ClassificationEvaluateResponse => {
+  const keys = Object.keys(arg);
+  return (
+    keys.length === 1 &&
+    keys[0] === ANALYSIS_CONFIG_TYPE.CLASSIFICATION &&
+    arg?.classification?.multiclass_confusion_matrix !== undefined
+  );
+};
+
 export interface DataFrameAnalyticsConfig {
   id: DataFrameAnalyticsId;
   // Description attribute is not supported yet
@@ -254,17 +304,14 @@ export function getValuesFromResponse(response: RegressionEvaluateResponse) {
 
   return { meanSquaredError, rSquared };
 }
-interface RegressionResultsSearchBoolQuery {
+interface ResultsSearchBoolQuery {
   bool: Dictionary<any>;
 }
-interface RegressionResultsSearchTermQuery {
+interface ResultsSearchTermQuery {
   term: Dictionary<any>;
 }
 
-export type RegressionResultsSearchQuery =
-  | RegressionResultsSearchBoolQuery
-  | RegressionResultsSearchTermQuery
-  | SavedSearchQuery;
+export type ResultsSearchQuery = ResultsSearchBoolQuery | ResultsSearchTermQuery | SavedSearchQuery;
 
 export function getEvalQueryBody({
   resultsField,
@@ -274,23 +321,44 @@ export function getEvalQueryBody({
 }: {
   resultsField: string;
   isTraining: boolean;
-  searchQuery?: RegressionResultsSearchQuery;
+  searchQuery?: ResultsSearchQuery;
   ignoreDefaultQuery?: boolean;
 }) {
-  let query: RegressionResultsSearchQuery = {
+  let query: ResultsSearchQuery = {
     term: { [`${resultsField}.is_training`]: { value: isTraining } },
   };
 
   if (searchQuery !== undefined && ignoreDefaultQuery === true) {
     query = searchQuery;
-  } else if (searchQuery !== undefined && isRegressionResultsSearchBoolQuery(searchQuery)) {
+  } else if (searchQuery !== undefined && isResultsSearchBoolQuery(searchQuery)) {
     const searchQueryClone = cloneDeep(searchQuery);
     searchQueryClone.bool.must.push(query);
     query = searchQueryClone;
   }
   return query;
 }
 
+interface EvaluateMetrics {
+  classification: {
+    multiclass_confusion_matrix: object;
+  };
+  regression: {
+    r_squared: object;
+    mean_squared_error: object;
+  };
+}
+
+interface LoadEvalDataConfig {
+  isTraining: boolean;
+  index: string;
+  dependentVariable: string;
+  resultsField: string;
+  predictionFieldName?: string;
+  searchQuery?: ResultsSearchQuery;
+  ignoreDefaultQuery?: boolean;
+  jobType: ANALYSIS_CONFIG_TYPE;
+}
+
 export const loadEvalData = async ({
   isTraining,
   index,
@@ -299,34 +367,38 @@ export const loadEvalData = async ({
   predictionFieldName,
   searchQuery,
   ignoreDefaultQuery,
-}: {
-  isTraining: boolean;
-  index: string;
-  dependentVariable: string;
-  resultsField: string;
-  predictionFieldName?: string;
-  searchQuery?: RegressionResultsSearchQuery;
-  ignoreDefaultQuery?: boolean;
-}) => {
+  jobType,
+}: LoadEvalDataConfig) => {
   const results: LoadEvaluateResult = { success: false, eval: null, error: null };
   const defaultPredictionField = `${dependentVariable}_prediction`;
-  const predictedField = `${resultsField}.${
+  let predictedField = `${resultsField}.${
     predictionFieldName ? predictionFieldName : defaultPredictionField
   }`;
 
+  if (jobType === ANALYSIS_CONFIG_TYPE.CLASSIFICATION) {
+    predictedField = `${predictedField}.keyword`;
+  }
+
   const query = getEvalQueryBody({ resultsField, isTraining, searchQuery, ignoreDefaultQuery });
 
+  const metrics: EvaluateMetrics = {
+    classification: {
+      multiclass_confusion_matrix: {},
+    },
+    regression: {
+      r_squared: {},
+      mean_squared_error: {},
+    },
+  };
+
   const config = {
     index,
     query,
     evaluation: {
-      regression: {
+      [jobType]: {
         actual_field: dependentVariable,
         predicted_field: predictedField,
-        metrics: {
-          r_squared: {},
-          mean_squared_error: {},
-        },
+        metrics: metrics[jobType as keyof EvaluateMetrics],
       },
     },
   };
@@ -341,3 +413,57 @@ export const loadEvalData = async ({
     return results;
   }
 };
+
+interface TrackTotalHitsSearchResponse {
+  hits: {
+    total: {
+      value: number;
+      relation: string;
+    };
+    hits: any[];
+  };
+}
+
+interface LoadDocsCountConfig {
+  ignoreDefaultQuery?: boolean;
+  isTraining: boolean;
+  searchQuery: SavedSearchQuery;
+  resultsField: string;
+  destIndex: string;
+}
+
+interface LoadDocsCountResponse {
+  docsCount: number | null;
+  success: boolean;
+}
+
+export const loadDocsCount = async ({
+  ignoreDefaultQuery = true,
+  isTraining,
+  searchQuery,
+  resultsField,
+  destIndex,
+}: LoadDocsCountConfig): Promise<LoadDocsCountResponse> => {
+  const query = getEvalQueryBody({ resultsField, isTraining, ignoreDefaultQuery, searchQuery });
+
+  try {
+    const body: SearchQuery = {
+      track_total_hits: true,
+      query,
+    };
+
+    const resp: TrackTotalHitsSearchResponse = await ml.esSearch({
+      index: destIndex,
+      size: 0,
+      body,
+    });
+
+    const docsCount = resp.hits.total && resp.hits.total.value;
+    return { docsCount, success: docsCount !== undefined };
+  } catch (e) {
+    return {
+      docsCount: null,
+      success: false,
+    };
+  }
+};

x-pack/legacy/plugins/ml/public/application/data_frame_analytics/common/fields.ts

@@ -77,7 +77,7 @@ export const sortRegressionResultsFields = (
 ) => {
   const dependentVariable = getDependentVar(jobConfig.analysis);
   const resultsField = jobConfig.dest.results_field;
-  const predictedField = getPredictedFieldName(resultsField, jobConfig.analysis);
+  const predictedField = getPredictedFieldName(resultsField, jobConfig.analysis, true);
   if (a === `${resultsField}.is_training`) {
     return -1;
   }
@@ -96,6 +96,14 @@ export const sortRegressionResultsFields = (
   if (b === dependentVariable) {
     return 1;
   }
+
+  if (a === `${resultsField}.prediction_probability`) {
+    return -1;
+  }
+  if (b === `${resultsField}.prediction_probability`) {
+    return 1;
+  }
+
   return a.localeCompare(b);
 };
 
@@ -107,7 +115,7 @@ export const sortRegressionResultsColumns = (
 ) => (a: string, b: string) => {
   const dependentVariable = getDependentVar(jobConfig.analysis);
   const resultsField = jobConfig.dest.results_field;
-  const predictedField = getPredictedFieldName(resultsField, jobConfig.analysis);
+  const predictedField = getPredictedFieldName(resultsField, jobConfig.analysis, true);
 
   const typeofA = typeof obj[a];
   const typeofB = typeof obj[b];
@@ -136,6 +144,14 @@ export const sortRegressionResultsColumns = (
     return 1;
   }
 
+  if (a === `${resultsField}.prediction_probability`) {
+    return -1;
+  }
+
+  if (b === `${resultsField}.prediction_probability`) {
+    return 1;
+  }
+
   if (typeofA !== 'string' && typeofB === 'string') {
     return 1;
   }
@@ -184,6 +200,43 @@ export function getFlattenedFields(obj: EsDocSource, resultsField: string): EsFi
   return flatDocFields.filter(f => f !== ML__ID_COPY);
 }
 
+export const getDefaultClassificationFields = (
+  docs: EsDoc[],
+  jobConfig: DataFrameAnalyticsConfig
+): EsFieldName[] => {
+  if (docs.length === 0) {
+    return [];
+  }
+  const resultsField = jobConfig.dest.results_field;
+  const newDocFields = getFlattenedFields(docs[0]._source, resultsField);
+  return newDocFields
+    .filter(k => {
+      if (k === `${resultsField}.is_training`) {
+        return true;
+      }
+      // predicted value of dependent variable
+      if (k === getPredictedFieldName(resultsField, jobConfig.analysis, true)) {
+        return true;
+      }
+      // actual value of dependent variable
+      if (k === getDependentVar(jobConfig.analysis)) {
+        return true;
+      }
+
+      if (k === `${resultsField}.prediction_probability`) {
+        return true;
+      }
+
+      if (k.split('.')[0] === resultsField) {
+        return false;
+      }
+
+      return docs.some(row => row._source[k] !== null);
+    })
+    .sort((a, b) => sortRegressionResultsFields(a, b, jobConfig))
+    .slice(0, DEFAULT_REGRESSION_COLUMNS);
+};
+
 export const getDefaultRegressionFields = (
   docs: EsDoc[],
   jobConfig: DataFrameAnalyticsConfig