feat: clean-up annoy metrics (#42)

vicinity/backends/annoy.py

@@ -17,17 +17,17 @@
 class AnnoyArgs(BaseArgs):
     dim: int = 0
     metric: str = "cosine"
+    internal_metric: str = "dot"
     trees: int = 100
     length: int | None = None
 
 
 class AnnoyBackend(AbstractBackend[AnnoyArgs]):
     argument_class = AnnoyArgs
-    supported_metrics = {Metric.COSINE, Metric.EUCLIDEAN, Metric.INNER_PRODUCT}
+    supported_metrics = {Metric.COSINE, Metric.EUCLIDEAN}
     inverse_metric_mapping = {
         Metric.COSINE: "dot",
         Metric.EUCLIDEAN: "euclidean",
-        Metric.INNER_PRODUCT: "dot",
     }
 
     def __init__(
@@ -56,18 +56,21 @@ def from_vectors(
         if metric_enum not in cls.supported_metrics:
             raise ValueError(f"Metric '{metric_enum.value}' is not supported by AnnoyBackend.")
 
-        metric = cls._map_metric_to_string(metric_enum)
+        metric_string = metric_enum.value
+        internal_metric = cls._map_metric_to_string(metric_enum)
 
-        if metric == "dot":
+        if metric_enum == Metric.COSINE:
             vectors = normalize(vectors)
 
         dim = vectors.shape[1]
-        index = AnnoyIndex(f=dim, metric=metric)  # type: ignore
+        index = AnnoyIndex(f=dim, metric=internal_metric)  # type: ignore
         for i, vector in enumerate(vectors):
             index.add_item(i, vector)
         index.build(trees)
 
-        arguments = AnnoyArgs(dim=dim, metric=metric, trees=trees, length=len(vectors))  # type: ignore
+        arguments = AnnoyArgs(
+            dim=dim, metric=metric_string, trees=trees, length=len(vectors), internal_metric=internal_metric
+        )  # type: ignore
         return AnnoyBackend(index, arguments=arguments)
 
     @property
@@ -89,7 +92,7 @@ def load(cls: type[AnnoyBackend], base_path: Path) -> AnnoyBackend:
         """Load the vectors from a path."""
         path = Path(base_path) / "index.bin"
         arguments = AnnoyArgs.load(base_path / "arguments.json")
-        index = AnnoyIndex(arguments.dim, arguments.metric)  # type: ignore
+        index = AnnoyIndex(arguments.dim, arguments.internal_metric)  # type: ignore
         index.load(str(path))
 
         return cls(index, arguments=arguments)
@@ -106,11 +109,11 @@ def query(self, vectors: npt.NDArray, k: int) -> QueryResult:
         """Query the backend."""
         out = []
         for vec in vectors:
-            if self.arguments.metric == "dot":
+            if self.arguments.metric == "cosine":
                 vec = normalize(vec)
             indices, scores = self.index.get_nns_by_vector(vec, k, include_distances=True)
             scores_array = np.asarray(scores)
-            if self.arguments.metric == "dot":
+            if self.arguments.metric == "cosine":
                 # Convert cosine similarity to cosine distance
                 scores_array = 1 - scores_array
             out.append((np.asarray(indices), scores_array))