This is an automated email from the ASF dual-hosted git repository. spmallette pushed a commit to branch TINKERPOP-3158 in repository https://gitbox.apache.org/repos/asf/tinkerpop.git
commit 133c82166270a2041f25bb8d6e17ad9a9ef78b5b Author: Stephen Mallette <[email protected]> AuthorDate: Thu May 8 13:19:53 2025 -0400 Cleaned up docs --- .../reference/implementations-tinkergraph.asciidoc | 48 +++++++++++++--------- 1 file changed, 28 insertions(+), 20 deletions(-) diff --git a/docs/src/reference/implementations-tinkergraph.asciidoc b/docs/src/reference/implementations-tinkergraph.asciidoc index daec3aa12b..76d53e0f21 100644 --- a/docs/src/reference/implementations-tinkergraph.asciidoc +++ b/docs/src/reference/implementations-tinkergraph.asciidoc @@ -306,10 +306,10 @@ g.addV("person").property("name", "Bob").property("embedding", new float[]{0.0f, g.addV("person").property("name", "Charlie").property("embedding", new float[]{0.0f, 0.0f, 1.0f}).iterate() g.addV("person").property("name", "Dave").property("embedding", new float[]{0.9f, 0.1f, 0.0f}).iterate() byElementParams = [key: "embedding", topK: 2] <4> -g.V().has("name", "Alice").call("tinker.search.vector.topKByElement", byElementParams).toList() <5> +g.V().has("name", "Alice").call("tinker.search.vector.topKByElement", byElementParams) <5> byElementParams = [key: "embedding", topK: 2, element: "vertex"] <6> embedding = new float[]{1.0f, 0.0f, 0.0f} -g.inject([embedding]).unfold().call("tinker.search.vector.topKByEmbedding", params).toList() <7> +g.inject([embedding]).unfold().call("tinker.search.vector.topKByEmbedding", params) <7> ---- <1> Register the vector search service for "topKByElement". @@ -325,20 +325,7 @@ The `call()` step returns a list of maps, each containing: * `distance`: The similarity score between the query vector and the result * `element`: The vertex or edge that matches the query -Vector indices can also be created for edges: - -[gremlin-groovy] ----- -graph.getServiceRegistry().registerService(new TinkerVectorSearchByElementFactory(graph)) -graph.createIndex(TinkerIndexType.VECTOR, "embedding", Edge.class, indexConfig) -alice = g.V().has("name", "Alice").next() -bob = g.V().has("name", "Bob").next() -charlie = g.V().has("name", "Charlie").next() -g.addE("knows").from(alice).to(bob).property("embedding", new float[]{1.0f, 0.0f, 0.0f}).iterate() -g.addE("knows").from(bob).to(charlie).property("embedding", new float[]{0.0f, 1.0f, 0.0f}).iterate() -params = [key: "embedding"] -g.E().has("embedding", new float[]{1.0f, 0.0f, 0.0f}).call("tinker.search.vector.topKByElement", params).toList() ----- +TIP: Vector indices can also be created for edges. TinkerGraph supports various distance functions for vector similarity search: @@ -352,9 +339,8 @@ TinkerGraph supports various distance functions for vector similarity search: You can specify the distance function when creating the vector index: -[gremlin-groovy] +[source,groovy] ---- -graph.getServiceRegistry().registerService(new TinkerVectorSearchByElementFactory(graph)) indexConfig = [dimension: 3, distanceType: TinkerIndexType.Vector.EUCLIDEAN] graph.createIndex(TinkerIndexType.VECTOR, "embedding", Vertex.class, indexConfig) ---- @@ -376,7 +362,7 @@ These options are specified when creating a vector index: Here's an example of creating a vector index with custom configuration options: -[gremlin-groovy] +[source,groovy] ---- graph.getServiceRegistry().registerService(new TinkerVectorSearchFactory(graph)) indexConfig = [ @@ -395,7 +381,29 @@ TIP: Constants for all the configuration values can be found in `TinkerVectorInd For example, "dimension" can be referenced as `TinkerVectorIndex.CONFIG_DIMENSION`. Note that the distance functions can be used directly with the `TinkerVectorDistanceFactory` service. It allows -calculation of the distance between the elements +calculation of the distance between the starting and ending elements in a `Path`. + +[gremlin-groovy] +---- +graph.getServiceRegistry().registerService(new TinkerVectorDistanceFactory(graph)) +g.addV("person").property("name", "Alice").property("embedding", new float[]{1.0f, 0.0f, 0.0f}).as('a'). + addV("person").property("name", "Bob").property("embedding", new float[]{0.0f, 1.0f, 0.0f}).as('b'). + addV("person").property("name", "Charlie").property("embedding", new float[]{0.0f, 0.0f, 1.0f}).as('c'). + addV("person").property("name", "Dave").property("embedding", new float[]{0.9f, 0.1f, 0.0f}).as('d'). + addE('knows').from('a').to('b'). + addE('knows').from('b').to('c'). + addE('knows').from('b').to('d'). + addE('knows').from('c').to('d').iterate() +params = [key: "embedding", distanceFunction: "COSINE"] +g.V().has('name','Alice').out().out().path() +g.V().has('name','Alice').out().out().path(). + project('path','distance'). + by(). + by(call('tinker.vector.distance', params)) +---- + +In the example above, the `Path` object is projected to a `Map` that has the `Path` itself and the vector distance +between the start and end vertex of the path. [[tinkergraph-gremlin-tx]] === Transactions
