Fix AzureDocumentDB score threshold for EuclideanDistance

MEVD/src/AzureDocumentDB/AzureDocumentDB.csproj

@@ -1,7 +1,7 @@
 <Project Sdk="Microsoft.NET.Sdk">
 
   <PropertyGroup>
-    <Version>1.0.0</Version>
+    <Version>1.0.1</Version>
     <AssemblyName>CommunityToolkit.VectorData.AzureDocumentDB</AssemblyName>
     <RootNamespace>$(AssemblyName)</RootNamespace>
     <TargetFrameworks>net10.0;net8.0;netstandard2.1;net472</TargetFrameworks>

MEVD/src/AzureDocumentDB/DocumentDBCollection.cs

@@ -406,7 +406,7 @@ _ when vectorProperty.EmbeddingGenerationDispatcher is not null
         // Add score threshold filter as a $match stage if specified
         if (options.ScoreThreshold.HasValue)
         {
-            pipeline.Add(DocumentDBCollectionSearchMapping.GetScoreThresholdMatchQuery(ScorePropertyName, options.ScoreThreshold.Value));
+            pipeline.Add(DocumentDBCollectionSearchMapping.GetScoreThresholdMatchQuery(ScorePropertyName, options.ScoreThreshold.Value, vectorProperty.DistanceFunction));
         }
 
         const string OperationName = "Aggregate";

MEVD/src/AzureDocumentDB/DocumentDBCollectionSearchMapping.cs

@@ -98,17 +98,27 @@ public static BsonDocument GetProjectionQuery(string scorePropertyName, string d
 
     /// <summary>Returns a $match stage to filter results by score threshold.</summary>
     /// <remarks>
-    /// Azure DocumentDB returns a similarity score where higher values mean more similar,
-    /// so we filter with $gte to keep results at or above the threshold.
+    /// For COS and IP Azure DocumentDB returns a similarity score where higher values mean more similar,
+    /// so we filter with $gte. For L2 it returns a distance where lower values mean more similar,
+    /// so we filter with $lte.
     /// </remarks>
-    public static BsonDocument GetScoreThresholdMatchQuery(string scorePropertyName, double scoreThreshold)
-        => new()
+    public static BsonDocument GetScoreThresholdMatchQuery(string scorePropertyName, double scoreThreshold, string? distanceFunction)
+    {
+        var comparisonOperator = GetVectorPropertyDistanceFunction(distanceFunction) switch
+        {
+            DistanceFunction.CosineDistance or DistanceFunction.DotProductSimilarity => "$gte",
+            DistanceFunction.EuclideanDistance => "$lte",
+            _ => throw new NotSupportedException($"Score threshold is not supported for distance function '{distanceFunction}'.")
+        };
+
+        return new()
         {
             {
                 "$match", new BsonDocument
                 {
-                    { scorePropertyName, new BsonDocument { { "$gte", scoreThreshold } } }
+                    { scorePropertyName, new BsonDocument { { comparisonOperator, scoreThreshold } } }
                 }
             }
         };
+    }
 }

MEVD/test/AzureDocumentDB.ConformanceTests/DocumentDBDistanceFunctionTests.cs

@@ -2,7 +2,6 @@
 // The .NET Foundation licenses this file to you under the MIT license.
 
 using AzureDocumentDB.ConformanceTests.Support;
-using Microsoft.Extensions.VectorData;
 using VectorData.ConformanceTests;
 using VectorData.ConformanceTests.Support;
 using Xunit;
@@ -18,11 +17,6 @@ public class DocumentDBDistanceFunctionTests(DocumentDBDistanceFunctionTests.Fix
     public override Task HammingDistance() => Assert.ThrowsAsync<NotSupportedException>(base.HammingDistance);
     public override Task ManhattanDistance() => Assert.ThrowsAsync<NotSupportedException>(base.ManhattanDistance);
 
-    // AzureDocumentDB EuclideanDistance doesn't correctly filter by score threshold (returns all results).
-    // See https://github.com/CommunityToolkit/AI/issues/6.
-    protected override Task TestScoreThreshold(VectorStoreCollection<int, DistanceFunctionTests<int>.SearchRecord> collection)
-        => Task.CompletedTask;
-
     public new class Fixture() : DistanceFunctionTests<int>.Fixture
     {
         public override TestStore TestStore => DocumentDBTestStore.Instance;