COH-32201 - Fix HNSWIndex implementation in Python Client

Author: dhirupandey

examples/vector_search.py

@@ -10,7 +10,7 @@
 from sentence_transformers import SentenceTransformer
 
 from coherence import NamedMap, Session
-from coherence.ai import FloatVector, QueryResult, SimilaritySearch, Vectors
+from coherence.ai import FloatVector, HnswIndex, QueryResult, SimilaritySearch, Vectors
 from coherence.extractor import Extractors, ValueExtractor
 from coherence.filter import Filter, Filters
 
@@ -100,6 +100,18 @@
 
 The SimilaritySearch aggregator is used to perform a Knn vector search on a
 cache in the same way that normal Coherence aggregators are used.
+
+HNSW Indexing
+=============
+
+Coherence includes an implementation of the HNSW index that can be used to
+speed up searches. The hierarchical navigable small world (HNSW) algorithm is
+a graph-based approximate nearest neighbor search technique.
+
+An index is added to a cache in Coherence by calling the add_index method on
+the cache. In this example, a HNSWIndex is created with a ValueExtractor that
+will extract the vector field from the cache value and an int parameter that
+specifies the number of dimensions the vector has.
 """
 
 
@@ -226,6 +238,11 @@ async def do_run() -> None:
         # NamedMap as a parameter
         movies_repo = MovieRepository(movie_db)
 
+        # To speed up the search, HNSW index is added to the cache using the
+        # VALUE_EXTRACTOR for the full plot vector and the dimensions of the
+        # vector.
+        await movie_db.add_index(HnswIndex(MovieRepository.VALUE_EXTRACTOR, MovieRepository.EMBEDDING_DIMENSIONS))
+
         # All of the movies data from filename  MOVIE_JSON_FILENAME is
         # processed and loaded into the movies_repo
         await movies_repo.load(MOVIE_JSON_FILENAME)

src/coherence/ai.py

@@ -392,7 +392,7 @@ class HnswIndex(AbstractEvolvable):
     def __init__(
         self,
         extractor: Union[ValueExtractor[T, E], str],
-        dimensions: int,
+        dimension: int,
         space_name: str = DEFAULT_SPACE_NAME,
         max_elements: int = DEFAULT_MAX_ELEMENTS,
         m: int = DEFAULT_M,
@@ -404,7 +404,7 @@ def __init__(
         Creates an instance of HnswIndex class.
 
         :param extractor: The ValueExtractor to use to extract the Vector.
-        :param dimensions: The number of dimensions in the vector.
+        :param dimension: The number of dimensions in the vector.
         :param space_name: The index space name.
         :param max_elements: The maximum number of elements the index can contain.
         :param m: The number of bidirectional links created for every new element during construction.
@@ -415,7 +415,7 @@ def __init__(
 
         super().__init__()
         self.extractor = extractor
-        self.dimensions = dimensions
+        self.dimension = dimension
         self.spaceName = space_name if space_name else ""
         self.maxElements = max_elements
         self.m = m

tests/unit/test_serialization.py

@@ -263,7 +263,7 @@ def test_HnswIndex_serialization() -> None:
     assert ser == (
         b'\x15{"@class": "coherence.hnsw.HnswIndex", "dataVersion": 0, '
         b'"binFuture": null, "extractor": {"@class": "extractor.UniversalExtractor", '
-        b'"name": "foo", "params": null}, "dimensions": 384, "spaceName": "COSINE", '
+        b'"name": "foo", "params": null}, "dimension": 384, "spaceName": "COSINE", '
         b'"maxElements": 4096, "m": 16, "efConstruction": 200, "efSearch": 50, '
         b'"randomSeed": 100}'
     )