中国移动云 ECloud ElasticSearch 向量搜索

中国移动 ECloud 向量搜索 是一款全托管的企业级分布式搜索和分析服务。中国移动 ECloud 向量搜索为结构化/非结构化数据提供低成本、高性能、可靠的检索和分析平台级产品服务。作为向量数据库，它支持多种索引类型和相似性距离方法。

您需要使用 `pip install -qU langchain-community` 安装 `langchain-community` 才能使用此集成

此笔记本展示了如何使用与 `ECloud ElasticSearch VectorStore` 相关的功能。要运行，您应该有一个正在运行的 中国移动 ECloud 向量搜索 实例

阅读 帮助文档 以快速熟悉和配置中国移动 ECloud ElasticSearch 实例。

实例运行后，请按照以下步骤拆分文档、获取嵌入、连接到百度云 Elasticsearch 实例、索引文档并执行向量检索。

#!pip install elasticsearch == 7.10.1

首先，我们要使用 `OpenAIEmbeddings`，因此我们必须获取 OpenAI API 密钥。

import getpass
import os

if "OPENAI_API_KEY" not in os.environ:
    os.environ["OPENAI_API_KEY"] = getpass.getpass("OpenAI API Key:")

其次，拆分文档并获取嵌入。

from langchain_community.document_loaders import TextLoader
from langchain_community.vectorstores import EcloudESVectorStore
from langchain_openai import OpenAIEmbeddings
from langchain_text_splitters import CharacterTextSplitter

API 参考：TextLoader | EcloudESVectorStore | OpenAIEmbeddings | CharacterTextSplitter

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)

embeddings = OpenAIEmbeddings()

ES_URL = "http://localhost:9200"
USER = "your user name"
PASSWORD = "your password"
indexname = "your index name"

然后，索引文档

docsearch = EcloudESVectorStore.from_documents(
    docs,
    embeddings,
    es_url=ES_URL,
    user=USER,
    password=PASSWORD,
    index_name=indexname,
    refresh_indices=True,
)

最后，查询和检索数据

query = "What did the president say about Ketanji Brown Jackson"
docs = docsearch.similarity_search(query, k=10)
print(docs[0].page_content)

一个常用的案例

def test_dense_float_vectore_lsh_cosine() -> None:
    """
    Test indexing with vectore type knn_dense_float_vector and  model-similarity of lsh-cosine
    this mapping is compatible with model of exact and similarity of l2/cosine
    this mapping is compatible with model of lsh and similarity of cosine
    """
    docsearch = EcloudESVectorStore.from_documents(
        docs,
        embeddings,
        es_url=ES_URL,
        user=USER,
        password=PASSWORD,
        index_name=indexname,
        refresh_indices=True,
        text_field="my_text",
        vector_field="my_vec",
        vector_type="knn_dense_float_vector",
        vector_params={"model": "lsh", "similarity": "cosine", "L": 99, "k": 1},
    )

    docs = docsearch.similarity_search(
        query,
        k=10,
        search_params={
            "model": "exact",
            "vector_field": "my_vec",
            "text_field": "my_text",
        },
    )
    print(docs[0].page_content)

    docs = docsearch.similarity_search(
        query,
        k=10,
        search_params={
            "model": "exact",
            "similarity": "l2",
            "vector_field": "my_vec",
            "text_field": "my_text",
        },
    )
    print(docs[0].page_content)

    docs = docsearch.similarity_search(
        query,
        k=10,
        search_params={
            "model": "exact",
            "similarity": "cosine",
            "vector_field": "my_vec",
            "text_field": "my_text",
        },
    )
    print(docs[0].page_content)

    docs = docsearch.similarity_search(
        query,
        k=10,
        search_params={
            "model": "lsh",
            "similarity": "cosine",
            "candidates": 10,
            "vector_field": "my_vec",
            "text_field": "my_text",
        },
    )
    print(docs[0].page_content)

带过滤器案例

def test_dense_float_vectore_exact_with_filter() -> None:
    """
    Test indexing with vectore type knn_dense_float_vector and default model/similarity
    this mapping is compatible with model of exact and similarity of l2/cosine
    """
    docsearch = EcloudESVectorStore.from_documents(
        docs,
        embeddings,
        es_url=ES_URL,
        user=USER,
        password=PASSWORD,
        index_name=indexname,
        refresh_indices=True,
        text_field="my_text",
        vector_field="my_vec",
        vector_type="knn_dense_float_vector",
    )
    # filter={"match_all": {}} ,default
    docs = docsearch.similarity_search(
        query,
        k=10,
        filter={"match_all": {}},
        search_params={
            "model": "exact",
            "vector_field": "my_vec",
            "text_field": "my_text",
        },
    )
    print(docs[0].page_content)

    # filter={"term": {"my_text": "Jackson"}}
    docs = docsearch.similarity_search(
        query,
        k=10,
        filter={"term": {"my_text": "Jackson"}},
        search_params={
            "model": "exact",
            "vector_field": "my_vec",
            "text_field": "my_text",
        },
    )
    print(docs[0].page_content)

    # filter={"term": {"my_text": "president"}}
    docs = docsearch.similarity_search(
        query,
        k=10,
        filter={"term": {"my_text": "president"}},
        search_params={
            "model": "exact",
            "similarity": "l2",
            "vector_field": "my_vec",
            "text_field": "my_text",
        },
    )
    print(docs[0].page_content)

相关

• 向量存储 概念指南
• 向量存储 操作指南