Improve presets for use cases (#322)

Signed-off-by: Tyler Ohlsen <[email protected]>
(cherry picked from commit ee56015)

common/constants.ts

@@ -153,6 +153,7 @@ export const SHARED_OPTIONAL_FIELDS = ['max_chunk_limit', 'description', 'tag'];
  */
 export const VECTOR_FIELD_PATTERN = `{{vector_field}}`;
 export const TEXT_FIELD_PATTERN = `{{text_field}}`;
+export const IMAGE_FIELD_PATTERN = `{{image_field}}`;
 export const QUERY_TEXT_PATTERN = `{{query_text}}`;
 export const QUERY_IMAGE_PATTERN = `{{query_image}}`;
 export const MODEL_ID_PATTERN = `{{model_id}}`;
@@ -163,7 +164,30 @@ export const FETCH_ALL_QUERY = {
   },
   size: 1000,
 };
-export const SEMANTIC_SEARCH_QUERY = {
+export const TERM_QUERY = {
+  query: {
+    term: {
+      [TEXT_FIELD_PATTERN]: {
+        value: QUERY_TEXT_PATTERN,
+      },
+    },
+  },
+};
+export const KNN_QUERY = {
+  _source: {
+    excludes: [VECTOR_FIELD_PATTERN],
+  },
+  query: {
+    knn: {
+      [VECTOR_FIELD_PATTERN]: {
+        vector: `{{vector}}`,
+      },
+      k: 10,
+      model_id: MODEL_ID_PATTERN,
+    },
+  },
+};
+export const SEMANTIC_SEARCH_QUERY_NEURAL = {
   _source: {
     excludes: [VECTOR_FIELD_PATTERN],
   },
@@ -177,7 +201,7 @@ export const SEMANTIC_SEARCH_QUERY = {
     },
   },
 };
-export const MULTIMODAL_SEARCH_QUERY = {
+export const MULTIMODAL_SEARCH_QUERY_NEURAL = {
   _source: {
     excludes: [VECTOR_FIELD_PATTERN],
   },
@@ -192,7 +216,25 @@ export const MULTIMODAL_SEARCH_QUERY = {
     },
   },
 };
-export const HYBRID_SEARCH_QUERY = {
+export const MULTIMODAL_SEARCH_QUERY_BOOL = {
+  query: {
+    bool: {
+      must: [
+        {
+          match: {
+            [TEXT_FIELD_PATTERN]: QUERY_TEXT_PATTERN,
+          },
+        },
+        {
+          match: {
+            [IMAGE_FIELD_PATTERN]: QUERY_IMAGE_PATTERN,
+          },
+        },
+      ],
+    },
+  },
+};
+export const HYBRID_SEARCH_QUERY_MATCH_NEURAL = {
   _source: {
     excludes: [VECTOR_FIELD_PATTERN],
   },
@@ -219,23 +261,64 @@ export const HYBRID_SEARCH_QUERY = {
     },
   },
 };
+export const HYBRID_SEARCH_QUERY_MATCH_TERM = {
+  _source: {
+    excludes: [VECTOR_FIELD_PATTERN],
+  },
+  query: {
+    hybrid: {
+      queries: [
+        {
+          match: {
+            [TEXT_FIELD_PATTERN]: {
+              query: QUERY_TEXT_PATTERN,
+            },
+          },
+        },
+        {
+          term: {
+            [TEXT_FIELD_PATTERN]: {
+              value: QUERY_TEXT_PATTERN,
+            },
+          },
+        },
+      ],
+    },
+  },
+};
 
 export const QUERY_PRESETS = [
   {
     name: 'Fetch all',
     query: customStringify(FETCH_ALL_QUERY),
   },
   {
-    name: WORKFLOW_TYPE.SEMANTIC_SEARCH,
-    query: customStringify(SEMANTIC_SEARCH_QUERY),
+    name: 'Term',
+    query: customStringify(TERM_QUERY),
+  },
+  {
+    name: 'Basic k-NN',
+    query: customStringify(KNN_QUERY),
+  },
+  {
+    name: `${WORKFLOW_TYPE.SEMANTIC_SEARCH} (neural)`,
+    query: customStringify(SEMANTIC_SEARCH_QUERY_NEURAL),
   },
   {
     name: WORKFLOW_TYPE.MULTIMODAL_SEARCH,
-    query: customStringify(MULTIMODAL_SEARCH_QUERY),
+    query: customStringify(MULTIMODAL_SEARCH_QUERY_BOOL),
+  },
+  {
+    name: `${WORKFLOW_TYPE.MULTIMODAL_SEARCH} (neural)`,
+    query: customStringify(MULTIMODAL_SEARCH_QUERY_NEURAL),
+  },
+  {
+    name: `Hybrid search (match & term queries)`,
+    query: customStringify(HYBRID_SEARCH_QUERY_MATCH_TERM),
   },
   {
-    name: WORKFLOW_TYPE.HYBRID_SEARCH,
-    query: customStringify(HYBRID_SEARCH_QUERY),
+    name: `Hybrid search (match & neural queries)`,
+    query: customStringify(HYBRID_SEARCH_QUERY_MATCH_NEURAL),
   },
 ] as QueryPreset[];
 

common/interfaces.ts

@@ -465,6 +465,7 @@ export type QuickConfigureFields = {
   embeddingModelId?: string;
   vectorField?: string;
   textField?: string;
+  imageField?: string;
   embeddingLength?: number;
 };
 

public/pages/workflows/new_workflow/quick_configure_inputs.tsx

@@ -124,6 +124,26 @@ export function QuickConfigureInputs(props: QuickConfigureInputsProps) {
               />
             </EuiCompressedFormRow>
             <EuiSpacer size="s" />
+            {props.workflowType === WORKFLOW_TYPE.MULTIMODAL_SEARCH && (
+              <>
+                <EuiCompressedFormRow
+                  label={'Image field'}
+                  isInvalid={false}
+                  helpText="The name of the document field containing the image binary"
+                >
+                  <EuiCompressedFieldText
+                    value={fieldValues?.imageField || ''}
+                    onChange={(e) => {
+                      setFieldValues({
+                        ...fieldValues,
+                        imageField: e.target.value,
+                      });
+                    }}
+                  />
+                </EuiCompressedFormRow>
+                <EuiSpacer size="s" />
+              </>
+            )}
             <EuiCompressedFormRow
               label={'Vector field'}
               isInvalid={false}

public/pages/workflows/new_workflow/quick_configure_modal.tsx

@@ -17,8 +17,11 @@ import {
   EuiCompressedFormRow,
 } from '@elastic/eui';
 import {
+  EMPTY_MAP_ENTRY,
+  IMAGE_FIELD_PATTERN,
   MODEL_ID_PATTERN,
   MapArrayFormValue,
+  MapFormValue,
   QuickConfigureFields,
   TEXT_FIELD_PATTERN,
   VECTOR_FIELD_PATTERN,
@@ -158,7 +161,8 @@ function injectQuickConfigureFields(
       // Semantic search / hybrid search: set defaults in the ingest processor, the index mappings,
       // and the preset query
       case WORKFLOW_TYPE.SEMANTIC_SEARCH:
-      case WORKFLOW_TYPE.HYBRID_SEARCH: {
+      case WORKFLOW_TYPE.HYBRID_SEARCH:
+      case WORKFLOW_TYPE.MULTIMODAL_SEARCH: {
         if (!isEmpty(quickConfigureFields) && workflow.ui_metadata?.config) {
           workflow.ui_metadata.config = updateIngestProcessorConfig(
             workflow.ui_metadata.config,
@@ -172,7 +176,12 @@ function injectQuickConfigureFields(
             workflow.ui_metadata.config,
             quickConfigureFields
           );
+          workflow.ui_metadata.config = updateSearchRequestProcessorConfig(
+            workflow.ui_metadata.config,
+            quickConfigureFields
+          );
         }
+        break;
       }
       case WORKFLOW_TYPE.CUSTOM:
       case undefined:
@@ -183,7 +192,7 @@ function injectQuickConfigureFields(
   return workflow;
 }
 
-// prefill ML ingest pipeline processor config, if applicable
+// prefill ML ingest processor config, if applicable
 function updateIngestProcessorConfig(
   config: WorkflowConfig,
   fields: QuickConfigureFields
@@ -192,10 +201,21 @@ function updateIngestProcessorConfig(
     if (field.id === 'model' && fields.embeddingModelId) {
       field.value = { id: fields.embeddingModelId };
     }
-    if (field.id === 'input_map' && fields.textField) {
-      field.value = [
-        [{ key: '', value: fields.textField }],
-      ] as MapArrayFormValue;
+    if (field.id === 'input_map' && (fields.textField || fields.imageField)) {
+      const inputMap = [] as MapFormValue;
+      if (fields.textField) {
+        inputMap.push({
+          key: '',
+          value: fields.textField,
+        });
+      }
+      if (fields.imageField) {
+        inputMap.push({
+          key: '',
+          value: fields.imageField,
+        });
+      }
+      field.value = [inputMap] as MapArrayFormValue;
     }
     if (field.id === 'output_map' && fields.vectorField) {
       field.value = [
@@ -207,6 +227,23 @@ function updateIngestProcessorConfig(
   return config;
 }
 
+// prefill ML search request processor config, if applicable
+function updateSearchRequestProcessorConfig(
+  config: WorkflowConfig,
+  fields: QuickConfigureFields
+): WorkflowConfig {
+  config.search.enrichRequest.processors[0].fields.forEach((field) => {
+    if (field.id === 'model' && fields.embeddingModelId) {
+      field.value = { id: fields.embeddingModelId };
+    }
+    if (field.id === 'input_map' || field.id === 'output_map') {
+      field.value = [[EMPTY_MAP_ENTRY]] as MapArrayFormValue;
+    }
+  });
+
+  return config;
+}
+
 // prefill index mappings/settings, if applicable
 function updateIndexConfig(
   config: WorkflowConfig,
@@ -226,6 +263,20 @@ function updateIndexConfig(
       },
     });
   }
+  if (fields.imageField) {
+    const existingMappings = JSON.parse(
+      config.ingest.index.mappings.value as string
+    );
+    config.ingest.index.mappings.value = customStringify({
+      ...existingMappings,
+      properties: {
+        ...(existingMappings.properties || {}),
+        [fields.imageField]: {
+          type: 'binary',
+        },
+      },
+    });
+  }
   if (fields.vectorField) {
     const existingMappings = JSON.parse(
       config.ingest.index.mappings.value as string
@@ -270,6 +321,13 @@ function updateSearchRequestConfig(
       fields.vectorField
     );
   }
+  if (fields.imageField) {
+    config.search.request.value = ((config.search.request.value ||
+      '') as string).replace(
+      new RegExp(IMAGE_FIELD_PATTERN, 'g'),
+      fields.imageField
+    );
+  }
 
   return config;
 }

public/pages/workflows/new_workflow/utils.ts

@@ -4,7 +4,11 @@
  */
 
 import { snakeCase } from 'lodash';
-import { MLIngestProcessor } from '../../../configs';
+import {
+  MLIngestProcessor,
+  MLSearchRequestProcessor,
+  NormalizationProcessor,
+} from '../../../configs';
 import {
   WorkflowTemplate,
   START_FROM_SCRATCH_WORKFLOW_NAME,
@@ -13,9 +17,8 @@ import {
   WORKFLOW_TYPE,
   FETCH_ALL_QUERY,
   customStringify,
-  SEMANTIC_SEARCH_QUERY,
-  MULTIMODAL_SEARCH_QUERY,
-  HYBRID_SEARCH_QUERY,
+  TERM_QUERY,
+  MULTIMODAL_SEARCH_QUERY_BOOL,
 } from '../../../../common';
 import { generateId } from '../../../utils';
 
@@ -125,9 +128,10 @@ function fetchSemanticSearchMetadata(): UIState {
   baseState.config.ingest.index.settings.value = customStringify({
     [`index.knn`]: true,
   });
-  baseState.config.search.request.value = customStringify(
-    SEMANTIC_SEARCH_QUERY
-  );
+  baseState.config.search.request.value = customStringify(TERM_QUERY);
+  baseState.config.search.enrichRequest.processors = [
+    new MLSearchRequestProcessor().toObj(),
+  ];
   return baseState;
 }
 
@@ -140,8 +144,11 @@ function fetchMultimodalSearchMetadata(): UIState {
     [`index.knn`]: true,
   });
   baseState.config.search.request.value = customStringify(
-    MULTIMODAL_SEARCH_QUERY
+    MULTIMODAL_SEARCH_QUERY_BOOL
   );
+  baseState.config.search.enrichRequest.processors = [
+    new MLSearchRequestProcessor().toObj(),
+  ];
   return baseState;
 }
 
@@ -153,7 +160,13 @@ function fetchHybridSearchMetadata(): UIState {
   baseState.config.ingest.index.settings.value = customStringify({
     [`index.knn`]: true,
   });
-  baseState.config.search.request.value = customStringify(HYBRID_SEARCH_QUERY);
+  baseState.config.search.request.value = customStringify(TERM_QUERY);
+  baseState.config.search.enrichResponse.processors = [
+    new NormalizationProcessor().toObj(),
+  ];
+  baseState.config.search.enrichRequest.processors = [
+    new MLSearchRequestProcessor().toObj(),
+  ];
   return baseState;
 }
 

server/resources/templates/hybrid_search.json

@@ -1,6 +1,6 @@
 {
     "name": "Hybrid Search",
-    "description": "A basic workflow containing the ingest pipeline and index configurations for performing hybrid search",
+    "description": "A basic workflow containing the ingest pipeline, search pipeline, and index configurations for performing hybrid search",
     "version": {
         "template": "1.0.0",
         "compatibility": [

server/resources/templates/multimodal_search.json

@@ -1,6 +1,6 @@
 {
     "name": "Multimodal Search",
-    "description": "A basic workflow containing the ingest pipeline and index configurations for performing multimodal search",
+    "description": "A basic workflow containing the ingest pipeline, search pipeline, and index configurations for performing multimodal search",
     "version": {
         "template": "1.0.0",
         "compatibility": [