Added sparse vector query test

Author: Mikep86

x-pack/plugin/rank-rrf/src/yamlRestTest/resources/rest-api-spec/test/linear/20_linear_retriever_simplified.yml

@@ -56,13 +56,15 @@ setup:
                 type: text
               timestamp:
                 type: date
-              vector:
+              dense_vector:
                 type: dense_vector
                 dims: 1
                 index: true
                 similarity: l2_norm
                 index_options:
                   type: flat
+              sparse_vector:
+                type: sparse_vector
 
   - do:
       bulk:
@@ -77,7 +79,10 @@ setup:
             "text_1": "foo match 1",
             "text_2": "x match 2",
             "timestamp": "2000-03-30",
-            "vector": [1]
+            "dense_vector": [1],
+            "sparse_vector": {
+              "foo": 1.0
+            }
           }
           {"index": {"_id": "2"}}
           {
@@ -87,7 +92,10 @@ setup:
             "text_1": "bar match 3",
             "text_2": "y match 4",
             "timestamp": "2010-02-08",
-            "vector": [2]
+            "dense_vector": [2],
+            "sparse_vector": {
+              "bar": 1.0
+            }
           }
           {"index": {"_id": "3"}}
           {
@@ -97,7 +105,10 @@ setup:
             "text_1": "baz match 5",
             "text_2": "z match 6",
             "timestamp": "2024-08-08",
-            "vector": [3]
+            "dense_vector": [3],
+            "sparse_vector": {
+              "baz": 1.0
+            }
           }
 
 ---
@@ -167,7 +178,7 @@ setup:
 ---
 "Semantic match per-field boosting using the simplified format":
   # The mock inference services generate synthetic vectors that don't accurately represent similarity to non-identical
-  # input, so it's hard to create a test produces intuitive results. Instead, we rely on the fact that the inference
+  # input, so it's hard to create a test that produces intuitive results. Instead, we rely on the fact that the inference
   # services generate consistent vectors (i.e. same input -> same output) to demonstrate that per-field boosting on
   # a semantic_text field can change the result order.
   - do:
@@ -255,20 +266,38 @@ setup:
   - close_to: { hits.hits.0._score: { value: 1.0, error: 0.0001 } }
 
 ---
-"Cannot query raw vector fields":
+"Can query sparse vector fields":
+  - do:
+      search:
+        index: test-index
+        body:
+          retriever:
+            linear:
+              fields: [ "sparse_vector" ]
+              query: "foo"
+              normalizer: "minmax"
+
+  # Lexical-only match, so max score is 1
+  - match: { hits.total.value: 1 }
+  - length: { hits.hits: 1 }
+  - match: { hits.hits.0._id: "1" }
+  - close_to: { hits.hits.0._score: { value: 1.0, error: 0.0001 } }
+
+---
+"Cannot query dense vector fields":
   - do:
       catch: bad_request
       search:
         index: test-index
         body:
           retriever:
             linear:
-              fields: [ "vector" ]
+              fields: [ "dense_vector" ]
               query: "foo"
               normalizer: "minmax"
 
   - contains: { error.root_cause.0.reason: "[linear] search failed - retrievers '[standard]' returned errors" }
-  - contains: { error.root_cause.0.suppressed.0.failed_shards.0.reason.reason: "Field [vector] of type [dense_vector] does not support match queries" }
+  - contains: { error.root_cause.0.suppressed.0.failed_shards.0.reason.reason: "Field [dense_vector] of type [dense_vector] does not support match queries" }
 
 ---
 "Filters are propagated":