cr

Hello there,

I noticed the `add_texts` method took a fair while for the Pinecone
VectorStore. This happened because embeddings were computed one by one.
This PR fixes it by calling `Embeddings.embed_documents` on all texts at
once.

I also took the liberty to harmonize initialization of embeddings to the
Embeddings class.

Cheers,

Olivier

langchain/vectorstores/pinecone.py

@@ -2,7 +2,8 @@
 from __future__ import annotations
 
 import uuid
-from typing import Any, Callable, Iterable, List, Optional, Tuple
+import warnings
+from typing import Any, Callable, Iterable, List, Optional, Tuple, Union
 
 from langchain.docstore.document import Document
 from langchain.embeddings.base import Embeddings
@@ -26,13 +27,13 @@ class Pinecone(VectorStore):
             pinecone.init(api_key="***", environment="...")
             index = pinecone.Index("langchain-demo")
             embeddings = OpenAIEmbeddings()
-            vectorstore = Pinecone(index, embeddings.embed_query, "text")
+            vectorstore = Pinecone(index, embeddings, "text")
     """
 
     def __init__(
         self,
         index: Any,
-        embedding_function: Callable,
+        embeddings: Union[Embeddings, Callable],
         text_key: str,
         namespace: Optional[str] = None,
     ):
@@ -50,7 +51,17 @@ class Pinecone(VectorStore):
                 f"got {type(index)}"
             )
         self._index = index
-        self._embedding_function = embedding_function
+        if isinstance(embeddings, Embeddings):
+            self._embeddings = embeddings
+        else:
+            # This is for backwards compatibility issues. Previously,
+            # embeddings.embed_query was passed in, not the whole class
+            warnings.warn(
+                "passing a function as embeddings is deprecated, "
+                "you should pass an Embedding object directly. "
+                "If this throws an error, that is why."
+            )
+            self._embeddings = embeddings.__self__  # type: ignore
         self._text_key = text_key
         self._namespace = namespace
 
@@ -78,13 +89,14 @@ class Pinecone(VectorStore):
         if namespace is None:
             namespace = self._namespace
         # Embed and create the documents
-        docs = []
-        ids = ids or [str(uuid.uuid4()) for _ in texts]
-        for i, text in enumerate(texts):
-            embedding = self._embedding_function(text)
-            metadata = metadatas[i] if metadatas else {}
+        _texts = list(texts)
+        ids = ids or [str(uuid.uuid4()) for _ in _texts]
+        embeddings = self._embeddings.embed_documents(_texts)
+        metadatas = metadatas or [{}] * len(_texts)
+        for metadata, text in zip(metadatas, _texts):
             metadata[self._text_key] = text
-            docs.append((ids[i], embedding, metadata))
+        docs = list(zip(ids, embeddings, metadatas))
+
         # upsert to Pinecone
         self._index.upsert(vectors=docs, namespace=namespace, batch_size=batch_size)
         return ids
@@ -109,7 +121,7 @@ class Pinecone(VectorStore):
         """
         if namespace is None:
             namespace = self._namespace
-        query_obj = self._embedding_function(query)
+        query_obj = self._embeddings.embed_query(query)
         docs = []
         results = self._index.query(
             [query_obj],
@@ -145,7 +157,7 @@ class Pinecone(VectorStore):
         """
         if namespace is None:
             namespace = self._namespace
-        query_obj = self._embedding_function(query)
+        query_obj = self._embeddings.embed_query(query)
         docs = []
         results = self._index.query(
             [query_obj],
@@ -244,7 +256,7 @@ class Pinecone(VectorStore):
 
             # upsert to Pinecone
             index.upsert(vectors=list(to_upsert), namespace=namespace)
-        return cls(index, embedding.embed_query, text_key, namespace)
+        return cls(index, embedding, text_key, namespace)
 
     @classmethod
     def from_existing_index(
@@ -263,6 +275,4 @@ class Pinecone(VectorStore):
                 "Please install it with `pip install pinecone-client`."
             )
 
-        return cls(
-            pinecone.Index(index_name), embedding.embed_query, text_key, namespace
-        )
+        return cls(pinecone.Index(index_name), embedding, text_key, namespace)