Move score_threshold to encoders

semantic_router/encoders/base.py

@@ -3,6 +3,7 @@ from pydantic import BaseModel, Field
 
 class BaseEncoder(BaseModel):
     name: str
+    score_threshold: float
     type: str = Field(default="base")
 
     class Config:

semantic_router/encoders/bm25.py

 from typing import Any
 
 from semantic_router.encoders import BaseEncoder
+from semantic_router.utils.logger import logger
 
 
 class BM25Encoder(BaseEncoder):
@@ -8,8 +9,8 @@ class BM25Encoder(BaseEncoder):
     idx_mapping: dict[int, int] | None = None
     type: str = "sparse"
 
-    def __init__(self, name: str = "bm25"):
-        super().__init__(name=name)
+    def __init__(self, name: str = "bm25", score_threshold: float = 0.82):
+        super().__init__(name=name, score_threshold=score_threshold)
         try:
             from pinecone_text.sparse import BM25Encoder as encoder
         except ImportError:
@@ -17,6 +18,7 @@ class BM25Encoder(BaseEncoder):
                 "Please install pinecone-text to use BM25Encoder. "
                 "You can install it with: `pip install semantic-router[hybrid]`"
             )
+        logger.info("Downloading and initializing BM25 model parameters.")
         self.model = encoder.default()
 
         params = self.model.get_params()

semantic_router/encoders/cohere.py

@@ -13,10 +13,11 @@ class CohereEncoder(BaseEncoder):
         self,
         name: str | None = None,
         cohere_api_key: str | None = None,
+        score_threshold: float = 0.3,
     ):
         if name is None:
             name = os.getenv("COHERE_MODEL_NAME", "embed-english-v3.0")
-        super().__init__(name=name)
+        super().__init__(name=name, score_threshold=score_threshold)
         cohere_api_key = cohere_api_key or os.getenv("COHERE_API_KEY")
         if cohere_api_key is None:
             raise ValueError("Cohere API key cannot be 'None'.")

semantic_router/encoders/fastembed.py

@@ -14,8 +14,10 @@ class FastEmbedEncoder(BaseEncoder):
     threads: Optional[int] = None
     _client: Any = PrivateAttr()
 
-    def __init__(self, **data):
-        super().__init__(**data)
+    def __init__(
+        self, score_threshold: float = 0.5, **data
+    ):  # TODO default score_threshold not thoroughly tested, should optimize
+        super().__init__(score_threshold=score_threshold, **data)
         self._client = self._initialize_client()
 
     def _initialize_client(self):

semantic_router/encoders/openai.py

@@ -17,10 +17,11 @@ class OpenAIEncoder(BaseEncoder):
         self,
         name: str | None = None,
         openai_api_key: str | None = None,
+        score_threshold: float = 0.82,
     ):
         if name is None:
             name = os.getenv("OPENAI_MODEL_NAME", "text-embedding-ada-002")
-        super().__init__(name=name)
+        super().__init__(name=name, score_threshold=score_threshold)
         api_key = openai_api_key or os.getenv("OPENAI_API_KEY")
         if api_key is None:
             raise ValueError("OpenAI API key cannot be 'None'.")

semantic_router/layer.py

@@ -7,8 +7,6 @@ import yaml
 from semantic_router.encoders import (
     BaseEncoder,
     CohereEncoder,
-    OpenAIEncoder,
-    FastEmbedEncoder,
 )
 from semantic_router.linear import similarity_matrix, top_scores
 from semantic_router.route import Route
@@ -153,27 +151,19 @@ class LayerConfig:
 class RouteLayer:
     index: np.ndarray | None = None
     categories: np.ndarray | None = None
-    score_threshold: float = 0.82
+    score_threshold: float
 
     def __init__(
-        self, encoder: BaseEncoder | None = None, routes: list[Route] | None = None
+        self,
+        encoder: BaseEncoder | None = None,
+        routes: list[Route] | None = None,
     ):
         logger.info("Initializing RouteLayer")
         self.index = None
         self.categories = None
         self.encoder = encoder if encoder is not None else CohereEncoder()
         self.routes: list[Route] = routes if routes is not None else []
-        # decide on default threshold based on encoder
-        # TODO move defaults to the encoder objects and extract from there
-        if isinstance(encoder, OpenAIEncoder):
-            self.score_threshold = 0.82
-        elif isinstance(encoder, CohereEncoder):
-            self.score_threshold = 0.3
-        elif isinstance(encoder, FastEmbedEncoder):
-            # TODO default not thoroughly tested, should optimize
-            self.score_threshold = 0.5
-        else:
-            self.score_threshold = 0.82
+        self.score_threshold = self.encoder.score_threshold
         # if routes list has been passed, we initialize index now
         if len(self.routes) > 0:
             # initialize index now

tests/unit/encoders/test_base.py

@@ -6,10 +6,11 @@ from semantic_router.encoders import BaseEncoder
 class TestBaseEncoder:
     @pytest.fixture
     def base_encoder(self):
-        return BaseEncoder(name="TestEncoder")
+        return BaseEncoder(name="TestEncoder", score_threshold=0.5)
 
     def test_base_encoder_initialization(self, base_encoder):
         assert base_encoder.name == "TestEncoder", "Initialization of name failed"
+        assert base_encoder.score_threshold == 0.5
 
     def test_base_encoder_call_method_not_implemented(self, base_encoder):
         with pytest.raises(NotImplementedError):

tests/unit/test_layer.py

@@ -69,7 +69,7 @@ routes:
 
 @pytest.fixture
 def base_encoder():
-    return BaseEncoder(name="test-encoder")
+    return BaseEncoder(name="test-encoder", score_threshold=0.5)
 
 
 @pytest.fixture
@@ -95,6 +95,7 @@ def routes():
 class TestRouteLayer:
     def test_initialization(self, openai_encoder, routes):
         route_layer = RouteLayer(encoder=openai_encoder, routes=routes)
+        assert openai_encoder.score_threshold == 0.82
         assert route_layer.score_threshold == 0.82
         assert len(route_layer.index) if route_layer.index is not None else 0 == 5
         assert (
@@ -105,9 +106,11 @@ class TestRouteLayer:
 
     def test_initialization_different_encoders(self, cohere_encoder, openai_encoder):
         route_layer_cohere = RouteLayer(encoder=cohere_encoder)
+        assert cohere_encoder.score_threshold == 0.3
         assert route_layer_cohere.score_threshold == 0.3
 
         route_layer_openai = RouteLayer(encoder=openai_encoder)
+        assert openai_encoder.score_threshold == 0.82
         assert route_layer_openai.score_threshold == 0.82
 
     def test_add_route(self, openai_encoder):
@@ -173,7 +176,7 @@ class TestRouteLayer:
 
     def test_failover_score_threshold(self, base_encoder):
         route_layer = RouteLayer(encoder=base_encoder)
-        assert route_layer.score_threshold == 0.82
+        assert route_layer.score_threshold == 0.5
 
     def test_json(self, openai_encoder, routes):
         with tempfile.NamedTemporaryFile(suffix=".yaml", delete=False) as temp: