Google Cloud Vertex Feature Store streamlines your ML feature management and online serving processes by letting you serve at low-latency your data in Google Cloud BigQuery, including the capacity to perform approximate neighbor retrieval for embeddings
This tutorial shows you how to easily perform low-latency vector search and approximate nearest neighbor retrieval directly from your BigQuery data, enabling powerful ML applications with minimal setup. We will do that using the VertexFSVectorStore class. This class is part of a set of 2 classes capable of providing a unified data storage and flexible vector search in Google Cloud:
  • BigQuery Vector Search: with BigQueryVectorStore class, which is ideal for rapid prototyping with no infrastructure setup and batch retrieval.
  • Feature Store Online Store: with VertexFSVectorStore class, enables low-latency retrieval with manual or scheduled data sync. Perfect for production-ready user-facing GenAI applications.
Diagram BQ-VertexFS

​
Getting started

​
Install the library

%pip install --upgrade --quiet  langchain langchain-google-vertexai "langchain-google-community[featurestore]"
To use the newly installed packages in this Jupyter runtime, you must restart the runtime. You can do this by running the cell below, which restarts the current kernel. 
import IPython

app = IPython.Application.instance()
app.kernel.do_shutdown(True)

​
Before you begin

​
Set your project ID

If you don’t know your project ID, try the following: 
PROJECT_ID = ""  # @param {type:"string"}

# Set the project id
! gcloud config set project {PROJECT_ID}

​
Set the region

You can also change the REGION variable used by BigQuery. Learn more about BigQuery regions. 
REGION = "us-central1"  # @param {type: "string"}

​
Set the dataset and table names

They will be your BigQuery Vector Store. 
DATASET = "my_langchain_dataset"  # @param {type: "string"}
TABLE = "doc_and_vectors"  # @param {type: "string"}

​
Authenticating your notebook environment

  • If you are using Colab to run this notebook, uncomment the cell below and continue.
  • If you are using Vertex AI Workbench, check out the setup instructions here.
# from google.colab import auth as google_auth

# google_auth.authenticate_user()

​
Demo: VertexFSVectorStore

​
Create an embedding class instance

You may need to enable Vertex AI API in your project by running gcloud services enable aiplatform.googleapis.com --project {PROJECT_ID} (replace {PROJECT_ID} with the name of your project). You can use any LangChain embeddings model. 
from langchain_google_vertexai import VertexAIEmbeddings

embedding = VertexAIEmbeddings(
    model_name="textembedding-gecko@latest", project=PROJECT_ID
)

​
Initialize VertexFSVectorStore

BigQuery Dataset and Table will be automatically created if they do not exist. See class definition here for all optional paremeters. 
from langchain_google_community import VertexFSVectorStore

store = VertexFSVectorStore(
    project_id=PROJECT_ID,
    dataset_name=DATASET,
    table_name=TABLE,
    location=REGION,
    embedding=embedding,
)

​
Add texts

Note: The first synchronization process will take around ~20 minutes because of Feature Online Store creation. 
all_texts = ["Apples and oranges", "Cars and airplanes", "Pineapple", "Train", "Banana"]
metadatas = [{"len": len(t)} for t in all_texts]

store.add_texts(all_texts, metadatas=metadatas)
You can also start a sync on demand by executing the sync_data method. 
store.sync_data()
When in a production environment, you can also use cron_schedule class parameter to setup an automatic scheduled synchronization. For example: 
store = VertexFSVectorStore(cron_schedule="TZ=America/Los_Angeles 00 13 11 8 *", ...)

​
Search for documents

query = "I'd like a fruit."
docs = store.similarity_search(query)
print(docs)

​
Search for documents by vector

query_vector = embedding.embed_query(query)
docs = store.similarity_search_by_vector(query_vector, k=2)
print(docs)

​
Search for documents with metadata filter

# This should only return "Banana" document.
docs = store.similarity_search_by_vector(query_vector, filter={"len": 6})
print(docs)

​
Add text with embeddings

You can also bring your own embeddings with theadd_texts_with_embeddings method. This is particularly useful for multimodal data which might require custom preprocessing before the embedding generation. 
items = ["some text"]
embs = embedding.embed(items)

ids = store.add_texts_with_embeddings(
    texts=["some text"], embs=embs, metadatas=[{"len": 1}]
)

​
Batch serving with BigQuery

You can simply use the method .to_bq_vector_store() to get a BigQueryVectorStore object, which offers optimized performances for batch use cases. All mandatory parameters will be automatically transferred from the existing class. See the class definition for all the parameters you can use. Moving back to BigQueryVectorStore is equivalently easy with the .to_vertex_fs_vector_store() method. 
store.to_bq_vector_store()  # pass optional VertexFSVectorStore parameters as arguments