VectorDistance: add new distance-type 'Geo', add vectorSearchCitiesGeo test #129

Author: Shubham

objectbox/CHANGELOG.md

@@ -1,5 +1,7 @@
 ## latest
 
+* Vector Search: You can now use the new `VectorDistanceType.GEO` distance-type to perform vector searches on
+  geographical coordinates. This is particularly useful for location-based applications.
 * Flutter for Linux/Windows, Dart Native: update to [objectbox-c 4.1.0](https://github.com/objectbox/objectbox-c/releases/tag/v4.1.0).
 * Flutter for Android: update to [objectbox-android 4.1.0](https://github.com/objectbox/objectbox-java/releases/tag/V4.1.0).
   If your project is [using Admin](https://docs.objectbox.io/data-browser#admin-for-android), make sure to

objectbox/example/dart-native/vectorsearch_cities/lib/model.dart

@@ -7,7 +7,7 @@ class City {
 
   String? name;
 
-  @HnswIndex(dimensions: 2)
+  @HnswIndex(dimensions: 2, distanceType: VectorDistanceType.geo)
   @Property(type: PropertyType.floatVector)
   List<double>? location;
 

objectbox/lib/src/annotations.dart

@@ -341,7 +341,18 @@ enum VectorDistanceType {
   /// The more negative the dot product, the higher the distance is (the farther the vectors are).
   ///
   /// Value range: 0.0 - 2.0 (nonlinear; 0.0: nearest, 1.0: orthogonal, 2.0: farthest)
-  dotProductNonNormalized
+  dotProductNonNormalized,
+
+  /// For geospatial coordinates aka latitude/longitude pairs.
+  ///
+  /// Note, that the vector dimension must be 2, with the latitude being the first element and longitude the second.
+  /// If the vector has more than 2 dimensions, the first 2 dimensions are used.
+  /// If the vector has fewer than 2 dimensions, the distance is zero.
+  ///
+  /// Internally, this uses haversine distance.
+  ///
+  /// Value range: 0 km - 6371 * π km (approx. 20015.09 km; half the Earth's circumference)
+  geo
 }
 
 /// Flags as a part of the [HnswIndex] configuration.

objectbox/lib/src/modelinfo/model_hnsw_params.dart

@@ -189,6 +189,8 @@ extension ModelVectorDistanceType on VectorDistanceType {
       return OBXVectorDistanceType.DotProduct;
     } else if (this == VectorDistanceType.dotProductNonNormalized) {
       return OBXVectorDistanceType.DotProductNonNormalized;
+    } else if (this == VectorDistanceType.geo) {
+      return OBXVectorDistanceType.Geo;
     } else {
       throw ArgumentError.value(this, "distanceType");
     }

objectbox_test/test/annotations_test.dart

@@ -22,6 +22,7 @@ void main() {
         OBXVectorDistanceType.DotProduct);
     expect(VectorDistanceType.dotProductNonNormalized.toConstant(),
         OBXVectorDistanceType.DotProductNonNormalized);
+    expect(VectorDistanceType.geo.toConstant(), OBXVectorDistanceType.Geo);
   });
 
   test("ModelHnswParams maps values", () {

objectbox_test/test/entity.dart

@@ -465,6 +465,10 @@ class HnswObject {
   @HnswIndex(dimensions: 2)
   List<double>? floatVector;
 
+  @Property(type: PropertyType.floatVector)
+  @HnswIndex(dimensions: 2, distanceType: VectorDistanceType.geo)
+  List<double>? floatVectorGeoCoordinates;
+
   final rel = ToOne<RelatedNamedEntity>();
 }
 

objectbox_test/test/hnsw_test.dart

@@ -72,6 +72,41 @@ void main() {
     expect(closest2.name, "node8");
   });
 
+  test('vectorSearchCitiesGeo', () {
+    // capital cities across Europe
+    List<String> cities = ["Berlin", "Paris", "Rome", "Madrid", "London"];
+    List<List<double>> coordinates = [
+      [52.5200, 13.4050],
+      [48.8566, 2.3522],
+      [41.9028, 12.4964],
+      [40.4168, -3.7038],
+      [51.5074, -0.1278]
+    ];
+
+    box.putMany(List.generate(cities.length, (i) {
+      return HnswObject()
+        ..name = cities[i]
+        ..floatVectorGeoCoordinates = coordinates[i];
+    }));
+
+    // lat/lng for Munich
+    final List<double> searchVector = [48.1371, 11.5754];
+
+    final query = box
+        .query(HnswObject_.floatVectorGeoCoordinates
+            .nearestNeighborsF32(searchVector, 5))
+        .build();
+    addTearDown(() => query.close());
+
+    final nearestCities = query.find();
+    expect(nearestCities.length, 5);
+    expect(nearestCities[0].name, "Berlin");
+    expect(nearestCities[1].name, "Paris");
+    expect(nearestCities[2].name, "Rome");
+    expect(nearestCities[3].name, "Madrid");
+    expect(nearestCities[4].name, "London");
+  });
+
   test('find offset limit', () {
     box.putMany(List.generate(15, (index) {
       final i = index + 1; // start at 1

objectbox_test/test/objectbox-model.json

@@ -667,7 +667,7 @@
     },
     {
       "id": "14:880388751413233760",
-      "lastPropertyId": "4:2308158275756661586",
+      "lastPropertyId": "5:1476994841170736323",
       "name": "HnswObject",
       "properties": [
         {
@@ -695,6 +695,13 @@
           "flags": 520,
           "indexId": "22:6527315700526716999",
           "relationTarget": "RelatedNamedEntity"
+        },
+        {
+          "id": "5:1476994841170736323",
+          "name": "floatVectorGeoCoordinates",
+          "type": 28,
+          "flags": 8,
+          "indexId": "23:6649884639373473085"
         }
       ],
       "relations": []
@@ -720,7 +727,7 @@
     }
   ],
   "lastEntityId": "15:4803284427984871569",
-  "lastIndexId": "22:6527315700526716999",
+  "lastIndexId": "23:6649884639373473085",
   "lastRelationId": "1:2155747579134420981",
   "lastSequenceId": "0:0",
   "modelVersion": 5,