Timescale Vector (Postgres)

Timescale Vector 是为 AI 应用而生的 PostgreSQL++。它使您能够有效地在 PostgreSQL 中存储和查询数十亿个向量嵌入。

PostgreSQL 也被称为 Postgres，是一个免费且开源的关系数据库管理系统 (RDBMS)，强调可扩展性和 SQL 兼容性。

本笔记本展示了如何使用 Postgres 向量数据库 (TimescaleVector) 执行自查询。在本笔记本中，我们将演示围绕 TimescaleVector 向量存储构建的 SelfQueryRetriever。

什么是 Timescale Vector？

Timescale Vector 是为 AI 应用而生的 PostgreSQL++。

Timescale Vector 使您能够有效地在 PostgreSQL 中存储和查询数百万个向量嵌入。

• 通过受 DiskANN 启发的索引算法，增强了 pgvector 在 10 亿+ 向量上更快、更准确的相似性搜索。
• 通过自动基于时间的的分区和索引，实现快速的基于时间的向量搜索。
• 提供熟悉的 SQL 界面，用于查询向量嵌入和关系数据。

Timescale Vector 是云 PostgreSQL，专为 AI 而生，可随您从 POC 扩展到生产

• 通过允许您在单个数据库中存储关系元数据、向量嵌入和时间序列数据，简化操作。
• 受益于坚如磐石的 PostgreSQL 基础，具有企业级功能，如流式备份和复制、高可用性和行级安全性。
• 通过企业级安全性和合规性，实现无忧体验。

如何访问 Timescale Vector

Timescale Vector 在云 PostgreSQL 平台 Timescale 上可用。（目前没有自托管版本。）

LangChain 用户可获得 Timescale Vector 的 90 天免费试用。

• 要开始使用，注册 Timescale，创建一个新数据库并按照本笔记本操作！
• 有关更多详细信息和性能基准，请参阅 Timescale Vector 解释博客。
• 有关在 python 中使用 Timescale Vector 的更多详细信息，请参阅安装说明。

创建 TimescaleVector 向量存储

首先，我们要创建一个 Timescale Vector 向量存储并用一些数据进行播种。我们创建了一个小的演示文档集，其中包含电影摘要。

注意：自查询检索器需要您安装 lark (pip install lark)。我们还需要 timescale-vector 包。

%pip install --upgrade --quiet  lark

%pip install --upgrade --quiet  timescale-vector

在本示例中，我们将使用 OpenAIEmbeddings，所以让我们加载您的 OpenAI API 密钥。

# Get openAI api key by reading local .env file
# The .env file should contain a line starting with `OPENAI_API_KEY=sk-`
import os

from dotenv import find_dotenv, load_dotenv

_ = load_dotenv(find_dotenv())

OPENAI_API_KEY = os.environ["OPENAI_API_KEY"]
# Alternatively, use getpass to enter the key in a prompt
# import os
# import getpass
# os.environ["OPENAI_API_KEY"] = getpass.getpass("OpenAI API Key:")

要连接到您的 PostgreSQL 数据库，您需要您的服务 URI，它可以在您创建新数据库后下载的速查表或 .env 文件中找到。

如果您还没有，注册 Timescale，并创建一个新数据库。

URI 看起来像这样：postgres://tsdbadmin:<password>@<id>.tsdb.cloud.timescale.com:<port>/tsdb?sslmode=require

# Get the service url by reading local .env file
# The .env file should contain a line starting with `TIMESCALE_SERVICE_URL=postgresql://`
_ = load_dotenv(find_dotenv())
TIMESCALE_SERVICE_URL = os.environ["TIMESCALE_SERVICE_URL"]

# Alternatively, use getpass to enter the key in a prompt
# import os
# import getpass
# TIMESCALE_SERVICE_URL = getpass.getpass("Timescale Service URL:")

from langchain_community.vectorstores.timescalevector import TimescaleVector
from langchain_core.documents import Document
from langchain_openai import OpenAIEmbeddings

embeddings = OpenAIEmbeddings()

API 参考：TimescaleVector | Document | OpenAIEmbeddings

这是我们将用于此演示的示例文档。数据是关于电影的，并且同时具有内容和元数据字段，其中包含有关特定电影的信息。

docs = [
    Document(
        page_content="A bunch of scientists bring back dinosaurs and mayhem breaks loose",
        metadata={"year": 1993, "rating": 7.7, "genre": "science fiction"},
    ),
    Document(
        page_content="Leo DiCaprio gets lost in a dream within a dream within a dream within a ...",
        metadata={"year": 2010, "director": "Christopher Nolan", "rating": 8.2},
    ),
    Document(
        page_content="A psychologist / detective gets lost in a series of dreams within dreams within dreams and Inception reused the idea",
        metadata={"year": 2006, "director": "Satoshi Kon", "rating": 8.6},
    ),
    Document(
        page_content="A bunch of normal-sized women are supremely wholesome and some men pine after them",
        metadata={"year": 2019, "director": "Greta Gerwig", "rating": 8.3},
    ),
    Document(
        page_content="Toys come alive and have a blast doing so",
        metadata={"year": 1995, "genre": "animated"},
    ),
    Document(
        page_content="Three men walk into the Zone, three men walk out of the Zone",
        metadata={
            "year": 1979,
            "director": "Andrei Tarkovsky",
            "genre": "science fiction",
            "rating": 9.9,
        },
    ),
]

最后，我们将创建我们的 Timescale Vector 向量存储。请注意，集合名称将是 PostgreSQL 表的名称，文档存储在该表中。

COLLECTION_NAME = "langchain_self_query_demo"
vectorstore = TimescaleVector.from_documents(
    embedding=embeddings,
    documents=docs,
    collection_name=COLLECTION_NAME,
    service_url=TIMESCALE_SERVICE_URL,
)

创建我们的自查询检索器

现在我们可以实例化我们的检索器。为此，我们需要预先提供一些关于我们的文档支持的元数据字段的信息，以及文档内容的简短描述。

from langchain.chains.query_constructor.schema import AttributeInfo
from langchain.retrievers.self_query.base import SelfQueryRetriever
from langchain_openai import OpenAI

# Give LLM info about the metadata fields
metadata_field_info = [
    AttributeInfo(
        name="genre",
        description="The genre of the movie",
        type="string or list[string]",
    ),
    AttributeInfo(
        name="year",
        description="The year the movie was released",
        type="integer",
    ),
    AttributeInfo(
        name="director",
        description="The name of the movie director",
        type="string",
    ),
    AttributeInfo(
        name="rating", description="A 1-10 rating for the movie", type="float"
    ),
]
document_content_description = "Brief summary of a movie"

# Instantiate the self-query retriever from an LLM
llm = OpenAI(temperature=0)
retriever = SelfQueryRetriever.from_llm(
    llm, vectorstore, document_content_description, metadata_field_info, verbose=True
)

API 参考：AttributeInfo | SelfQueryRetriever | OpenAI

使用 Timescale Vector 进行自查询检索

现在我们可以尝试实际使用我们的检索器了！

运行以下查询，并注意您如何在自然语言中指定查询、过滤器、复合过滤器（带有 AND、OR 的过滤器），自查询检索器将把该查询翻译成 SQL，并在 Timescale Vector (Postgres) 向量存储上执行搜索。

这说明了自查询检索器的强大功能。您可以使用它对您的向量存储执行复杂搜索，而无需您或您的用户直接编写任何 SQL！

# This example only specifies a relevant query
retriever.invoke("What are some movies about dinosaurs")

/Users/avtharsewrathan/sideprojects2023/timescaleai/tsv-langchain/langchain/libs/langchain/langchain/chains/llm.py:275: UserWarning: The predict_and_parse method is deprecated, instead pass an output parser directly to LLMChain.
  warnings.warn(
``````output
query='dinosaur' filter=None limit=None

[Document(page_content='A bunch of scientists bring back dinosaurs and mayhem breaks loose', metadata={'year': 1993, 'genre': 'science fiction', 'rating': 7.7}),
 Document(page_content='A bunch of scientists bring back dinosaurs and mayhem breaks loose', metadata={'year': 1993, 'genre': 'science fiction', 'rating': 7.7}),
 Document(page_content='Toys come alive and have a blast doing so', metadata={'year': 1995, 'genre': 'animated'}),
 Document(page_content='Toys come alive and have a blast doing so', metadata={'year': 1995, 'genre': 'animated'})]

# This example only specifies a filter
retriever.invoke("I want to watch a movie rated higher than 8.5")

query=' ' filter=Comparison(comparator=<Comparator.GT: 'gt'>, attribute='rating', value=8.5) limit=None

[Document(page_content='Three men walk into the Zone, three men walk out of the Zone', metadata={'year': 1979, 'genre': 'science fiction', 'rating': 9.9, 'director': 'Andrei Tarkovsky'}),
 Document(page_content='Three men walk into the Zone, three men walk out of the Zone', metadata={'year': 1979, 'genre': 'science fiction', 'rating': 9.9, 'director': 'Andrei Tarkovsky'}),
 Document(page_content='A psychologist / detective gets lost in a series of dreams within dreams within dreams and Inception reused the idea', metadata={'year': 2006, 'rating': 8.6, 'director': 'Satoshi Kon'}),
 Document(page_content='A psychologist / detective gets lost in a series of dreams within dreams within dreams and Inception reused the idea', metadata={'year': 2006, 'rating': 8.6, 'director': 'Satoshi Kon'})]

# This example specifies a query and a filter
retriever.invoke("Has Greta Gerwig directed any movies about women")

query='women' filter=Comparison(comparator=<Comparator.EQ: 'eq'>, attribute='director', value='Greta Gerwig') limit=None

[Document(page_content='A bunch of normal-sized women are supremely wholesome and some men pine after them', metadata={'year': 2019, 'rating': 8.3, 'director': 'Greta Gerwig'}),
 Document(page_content='A bunch of normal-sized women are supremely wholesome and some men pine after them', metadata={'year': 2019, 'rating': 8.3, 'director': 'Greta Gerwig'})]

# This example specifies a composite filter
retriever.invoke("What's a highly rated (above 8.5) science fiction film?")

query=' ' filter=Operation(operator=<Operator.AND: 'and'>, arguments=[Comparison(comparator=<Comparator.GTE: 'gte'>, attribute='rating', value=8.5), Comparison(comparator=<Comparator.EQ: 'eq'>, attribute='genre', value='science fiction')]) limit=None

[Document(page_content='Three men walk into the Zone, three men walk out of the Zone', metadata={'year': 1979, 'genre': 'science fiction', 'rating': 9.9, 'director': 'Andrei Tarkovsky'}),
 Document(page_content='Three men walk into the Zone, three men walk out of the Zone', metadata={'year': 1979, 'genre': 'science fiction', 'rating': 9.9, 'director': 'Andrei Tarkovsky'})]

# This example specifies a query and composite filter
retriever.invoke(
    "What's a movie after 1990 but before 2005 that's all about toys, and preferably is animated"
)

query='toys' filter=Operation(operator=<Operator.AND: 'and'>, arguments=[Comparison(comparator=<Comparator.GT: 'gt'>, attribute='year', value=1990), Comparison(comparator=<Comparator.LT: 'lt'>, attribute='year', value=2005), Comparison(comparator=<Comparator.EQ: 'eq'>, attribute='genre', value='animated')]) limit=None

[Document(page_content='Toys come alive and have a blast doing so', metadata={'year': 1995, 'genre': 'animated'})]

过滤器 k

我们还可以使用自查询检索器来指定 k：要获取的文档数量。

我们可以通过将 enable_limit=True 传递给构造函数来做到这一点。

retriever = SelfQueryRetriever.from_llm(
    llm,
    vectorstore,
    document_content_description,
    metadata_field_info,
    enable_limit=True,
    verbose=True,
)

# This example specifies a query with a LIMIT value
retriever.invoke("what are two movies about dinosaurs")

query='dinosaur' filter=None limit=2

[Document(page_content='A bunch of scientists bring back dinosaurs and mayhem breaks loose', metadata={'year': 1993, 'genre': 'science fiction', 'rating': 7.7}),
 Document(page_content='A bunch of scientists bring back dinosaurs and mayhem breaks loose', metadata={'year': 1993, 'genre': 'science fiction', 'rating': 7.7})]