feat: Add the PGVectorStore

README.md

@@ -9,7 +9,7 @@
 
 The `langchain-postgres` package implementations of core LangChain abstractions using `Postgres`.
 
-The package is released under the MIT license. 
+The package is released under the MIT license.
 
 Feel free to use the abstraction as provided or else modify them / extend them as appropriate for your own application.
 
@@ -23,22 +23,65 @@ The package currently only supports the [psycogp3](https://www.psycopg.org/psyco
 pip install -U langchain-postgres
 ```
 
-## Change Log
+## Usage
 
-0.0.6: 
-- Remove langgraph as a dependency as it was causing dependency conflicts.
-- Base interface for checkpointer changed in langgraph, so existing implementation would've broken regardless.
+### Vectorstore
 
-## Usage
+> [!NOTE]
+> See example for the [PGVector vectorstore here](https://github.com/langchain-ai/langchain-postgres/blob/main/examples/vectorstore.ipynb)
+`PGVector` is being deprecated. Please migrate to `PGVectorStore`.
+`PGVectorStore` is used for improved performance and manageability.
+See the [migration guide](https://github.com/langchain-ai/langchain-postgres/blob/main/examples/migrate_pgvector_to_pgvectorstore.md) for details on how to migrate from `PGVector` to `PGVectorStore`.
+
+> [!TIP]
+> All synchronous functions have corresponding asynchronous functions
+
+```python
+from langchain_postgres import PGEngine, PGVectorStore
+from langchain_core.embeddings import DeterministicFakeEmbedding
+import uuid
+
+# Replace these variable values
+engine = PGEngine.from_connection_string(url=CONNECTION_STRING)
+
+VECTOR_SIZE = 768
+embedding = DeterministicFakeEmbedding(size=VECTOR_SIZE)
+
+engine.init_vectorstore_table(
+    table_name="destination_table",
+    vector_size=VECTOR_SIZE,
+)
+
+store = PGVectorStore.create_sync(
+    engine=engine,
+    table_name=TABLE_NAME,
+    embedding_service=embedding,
+)
+
+all_texts = ["Apples and oranges", "Cars and airplanes", "Pineapple", "Train", "Banana"]
+metadatas = [{"len": len(t)} for t in all_texts]
+ids = [str(uuid.uuid4()) for _ in all_texts]
+docs = [
+    Document(id=ids[i], page_content=all_texts[i], metadata=metadatas[i]) for i in range(len(all_texts))
+]
+
+store.add_documents(docs)
+
+query = "I'd like a fruit."
+docs = store.similarity_search(query)
+print(docs)
+```
+
+For a detailed example on `PGVectorStore` see [here](https://github.com/langchain-ai/langchain-postgres/blob/main/examples/pg_vectorstore.ipynb).
 
 ### ChatMessageHistory
 
-The chat message history abstraction helps to persist chat message history 
+The chat message history abstraction helps to persist chat message history
 in a postgres table.
 
 PostgresChatMessageHistory is parameterized using a `table_name` and a `session_id`.
 
-The `table_name` is the name of the table in the database where 
+The `table_name` is the name of the table in the database where
 the chat messages will be stored.
 
 The `session_id` is a unique identifier for the chat session. It can be assigned
@@ -79,7 +122,6 @@ chat_history.add_messages([
 print(chat_history.messages)
 ```
 
-
 ### Vectorstore
 
 See example for the [PGVector vectorstore here](https://github.com/langchain-ai/langchain-postgres/blob/main/examples/vectorstore.ipynb)

examples/migrate_pgvector_to_pgvectorstore.md

@@ -0,0 +1,174 @@
+# Migrate a `PGVector` vector store to `PGVectorStore`
+
+This guide shows how to migrate from the [`PGVector`](https://github.com/langchain-ai/langchain-postgres/blob/main/langchain_postgres/vectorstores.py) vector store class to the [`PGVectorStore`](https://github.com/langchain-ai/langchain-postgres/blob/main/langchain_postgres/vectorstore.py) class.
+
+## Why migrate?
+
+This guide explains how to migrate your vector data from a PGVector-style database (two tables) to an PGVectoStore-style database (one table per collection) for improved performance and manageability.
+
+Migrating to the PGVectorStore interface provides the following benefits:
+
+- **Simplified management**: A single table contains data corresponding to a single collection, making it easier to query, update, and maintain.
+- **Improved metadata handling**: It stores metadata in columns instead of JSON, resulting in significant performance improvements.
+- **Schema flexibility**: The interface allows users to add tables into any database schema.
+- **Improved performance**: The single-table schema can lead to faster query execution, especially for large collections.
+- **Clear separation**: Clearly separate table and extension creation, allowing for distinct permissions and streamlined workflows.
+- **Secure Connections:** The PGVectorStore interface creates a secure connection pool that can be easily shared across your application using the `engine` object.
+
+## Migration process
+
+> **_NOTE:_**  The langchain-core library is installed to use the Fake embeddings service. To use a different embedding service, you'll need to install the appropriate library for your chosen provider. Choose embeddings services from [LangChain's Embedding models](https://python.langchain.com/v0.2/docs/integrations/text_embedding/).
+
+While you can use the  existing PGVector database, we **strongly recommend** migrating your data to the PGVectorStore-style schema to take full advantage of the performance benefits.
+
+### (Recommended) Data migration
+
+1. **Create a PG engine.**
+
+    ```python
+    from langchain_postgres import PGEngine
+
+    # Replace these variable values
+    engine = PGEngine.from_connection_string(url=CONNECTION_STRING)
+    ```
+
+    > **_NOTE:_** All sync methods have corresponding async methods.
+
+2. **Create a new table to migrate existing data.**
+
+    ```python
+    # Vertex AI embeddings uses a vector size of 768.
+    # Adjust this according to your embeddings service.
+    VECTOR_SIZE = 768
+
+    engine.init_vectorstore_table(
+        table_name="destination_table",
+        vector_size=VECTOR_SIZE,
+    )
+    ```
+
+    **(Optional) Customize your table.**
+
+    When creating your vectorstore table, you have the flexibility to define custom metadata and ID columns. This is particularly useful for:
+
+    - **Filtering**: Metadata columns allow you to easily filter your data within the vectorstore. For example, you might store the document source, date, or author as metadata for efficient retrieval.
+    - **Non-UUID Identifiers**: By default, the id_column uses UUIDs. If you need to use a different type of ID (e.g., an integer or string), you can define a custom id_column.
+
+    ```python
+    metadata_columns = [
+        Column(f"col_0_{collection_name}", "VARCHAR"),
+        Column(f"col_1_{collection_name}", "VARCHAR"),
+    ]
+    engine.init_vectorstore_table(
+        table_name="destination_table",
+        vector_size=VECTOR_SIZE,
+        metadata_columns=metadata_columns,
+        id_column=Column("langchain_id", "VARCHAR"),
+    )
+    ```
+
+3. **Create a vector store object to interact with the new data.**
+
+    > **_NOTE:_** The `FakeEmbeddings` embedding service is only used to initialise a vector store object, not to generate any embeddings. The embeddings are copied directly from the PGVector table.
+
+    ```python
+    from langchain_postgres import PGVectorStore
+    from langchain_core.embeddings import FakeEmbeddings
+
+    destination_vector_store = PGVectorStore.create_sync(
+        engine,
+        embedding_service=FakeEmbeddings(size=VECTOR_SIZE),
+        table_name="destination_table",
+    )
+    ```
+
+    If you have any customisations on the metadata or the id columns, add them to the vector store as follows:
+
+    ```python
+    from langchain_postgres import PGVectorStore
+    from langchain_core.embeddings import FakeEmbeddings
+
+    destination_vector_store = PGVectorStore.create_sync(
+        engine,
+        embedding_service=FakeEmbeddings(size=VECTOR_SIZE),
+        table_name="destination_table",
+        metadata_columns=[col.name for col in metadata_columns],
+        id_column="langchain_id",
+    )
+    ```
+
+4. **Migrate the data to the new table.**
+
+    ```python
+    from langchain_postgres.utils.pgvector_migrator import amigrate_pgvector_collection
+
+    migrate_pgvector_collection(
+        engine,
+        # Set collection name here
+        collection_name="collection_name",
+        vector_store=destination_vector_store,
+        # This deletes data from the original table upon migration. You can choose to turn it off.
+        delete_pg_collection=True,
+    )
+    ```
+
+    The data will only be deleted from the original table once all of it has been successfully copied to the destination table.
+
+> **TIP:** If you would like to migrate multiple collections, you can use the `alist_pgvector_collection_names` method to get the names of all collections, allowing you to iterate through them.
+>
+> ```python
+> from langchain_postgres.utils.pgvector_migrator import alist_pgvector_collection_names
+>
+> all_collection_names = list_pgvector_collection_names(engine)
+> print(all_collection_names)
+> ```
+
+### (Not Recommended) Use PGVectorStore interface on PGVector databases
+
+If you choose not to migrate your data, you can still use the PGVectorStore interface with your existing PGVector database. However, you won't benefit from the performance improvements of the PGVectorStore-style schema.
+
+1. **Create an PGVectorStore engine.**
+
+    ```python
+    from langchain_postgres import PGEngine
+
+    # Replace these variable values
+    engine = PGEngine.from_connection_string(url=CONNECTION_STRING)
+    ```
+
+    > **_NOTE:_** All sync methods have corresponding async methods.
+
+2. **Create a vector store object to interact with the data.**
+
+    Use the embeddings service used by your database. See [langchain docs](https://python.langchain.com/docs/integrations/text_embedding/) for reference.
+
+    ```python
+    from langchain_postgres import PGVectorStore
+    from langchain_core.embeddings import FakeEmbeddings
+
+    vector_store = PGVectorStore.create_sync(
+        engine=engine,
+        table_name="langchain_pg_embedding",
+        embedding_service=FakeEmbeddings(size=VECTOR_SIZE),
+        content_column="document",
+        metadata_json_column="cmetadata",
+        metadata_columns=["collection_id"],
+        id_column="id",
+    )
+    ```
+
+3. **Perform similarity search.**
+
+    Filter by collection id:
+
+    ```python
+    vector_store.similarity_search("query", k=5, filter=f"collection_id='{uuid}'")
+    ```
+
+    Filter by collection id and metadata:
+
+    ```python
+    vector_store.similarity_search(
+        "query", k=5, filter=f"collection_id='{uuid}' and cmetadata->>'col_name' = 'value'"
+    )
+    ```