Setup#

  • Create a collection with vector enabled
  • Create an index

Let create a collection

create-oass-collection

Then create a index

create-oass-collection

Then install dependencies, here is requirements.txt

boto3==1.26.25
boto3-stubs==1.24.26
botocore==1.29.25
botocore-stubs==1.27.42.post1
requests-aws4auth==1.1.2
opensearch-py==2.0.0

Client#

Let creat a opensearch client which using oass service

from opensearchpy import OpenSearch, RequestsHttpConnection
from requests_aws4auth import AWS4Auth
import boto3


service = 'aoss'
region = 'us-east-1'
credentials = boto3.Session().get_credentials()


# authentication
awsauth = AWS4Auth(
    credentials.access_key,
    credentials.secret_key,
    region,
    service,
    session_token=credentials.token,
)


# oass client
client = OpenSearch(
    hosts=[{"host": os.environ['OASS_URL'], "port": 443}],
    http_auth=awsauth,
    use_ssl=True,
    verify_certs=True,
    connection_class=RequestsHttpConnection,
    timeout=300,
)

Please note OASS_URL does not conttain https

Operations#

Let create a index which support KNN

# create index with vector support
client.indices.create(
  index='demo',
  body={
   "settings": {
      "index.knn": True
   },
   "mappings": {
      "properties": {
         "housing-vector": {
            "type": "knn_vector",
            "dimension": 3072
         },
         "title": {
            "type": "text"
         },
         "price": {
            "type": "long"
         },
         "location": {
            "type": "geo_point"
         }
      }
   }
}
)

Let index a simple data

# index data
client.index(
  index='demo',
  body={
  "housing-vector": e1,
  "title": "2 bedroom in downtown Seattle",
  "price": "2800",
  "location": "47.71, 122.00"
}
)

Then query data

# query index
client.search(
  index='demo',
  body={
    "size": 5,
    "query": {
        "knn": {
            "housing-vector": {
                "vector": e1,
                "k": 5
            }
        }
    }
}
)

List Documents#

Let list all documents

# list all document
response = client.search(
  index="demo",
  body={
    "size": 15,
    "query": {
      "match_all": {}
    }
  }
)


ids = [hit["_id"] for hit in response["hits"]["hits"]]

Delete all documents

for id in ids:
  client.delete(
    index="demo",
    id=id
  )

Get an document by id

client.get(
  index='demo',
  id='1%3A0%3AZa2hDo4BqbiA4twioVwz'
)

LangChain#

Let integrate with LangChain. Basically, it pipe user prompt => vector database => prompt model => response.

docsearch = OpenSearchVectorSearch(
  embedding_function=embeddings,
  opensearch_url=f"https://{os.environ['OASS_URL']}",
  http_auth=awsauth,
  timeout=300,
  use_ssl=True,
  verify_certs=True,
  connection_class=RequestsHttpConnection,
  index_name='demo',
  engine='nmslib'
  )

Then search on the existing document

docsearch.similarity_search(
  query='housing',
  vector_field='housing-vector',
  text_field='title',
  k=5
)

Search by vector

docsearch.similarity_search_by_vector(
  embedding=e1,
  k=5
)

Embedding#

Let use OpenAIEmbeddings to create sample embedding vectors.

First load OpenAI token

import os
from dotenv import load_dotenv
load_dotenv(".demo.env")

Then create a embedding model

from langchain.text_splitter import CharacterTextSplitter
from langchain_community.vectorstores import OpenSearchVectorSearch
from langchain_openai import OpenAIEmbeddings


# load txt
from langchain_community.document_loaders import TextLoader


loader = TextLoader("state_of_the_union.txt")
documents = loader.load()
text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0)
docs = text_splitter.split_documents(documents)


# create fist embedding vector
e1 = embeddings.embed_query(docs[0].page_content)

Here is the sample data state_of_the_union.txt

Reference#