跳到主要內容

DocArray

DocArray 是一款多功能、開放原始碼工具，用於管理您的多模態資料。它讓您可以隨意塑造您的資料，並提供彈性來使用各種文件索引後端儲存和搜尋資料。此外，它變得更好了 - 您可以利用您的 DocArray 文件索引來建立 DocArrayRetriever，並建構出色的 Langchain 應用程式！

本筆記本分為兩個部分。第一部分 介紹所有五個支援的文件索引後端。它提供有關設定和索引每個後端的指南，並指導您如何建構 DocArrayRetriever 以尋找相關文件。在第二部分中，我們將選擇其中一個後端，並透過基本範例來說明如何使用它。

文件索引後端​

import random

from docarray import BaseDoc
from docarray.typing import NdArray
from langchain_community.embeddings import FakeEmbeddings
from langchain_community.retrievers import DocArrayRetriever

embeddings = FakeEmbeddings(size=32)

在您開始建立索引之前，定義您的文件結構描述非常重要。這決定了您的文件將具有哪些欄位，以及每個欄位將保存的資料類型。

對於此示範，我們將建立一個有點隨機的結構描述，其中包含 'title' (str)、'title_embedding' (numpy 陣列)、'year' (int) 和 'color' (str)

class MyDoc(BaseDoc):
    title: str
    title_embedding: NdArray[32]
    year: int
    color: str

InMemoryExactNNIndex​

InMemoryExactNNIndex 將所有文件儲存在記憶體中。對於小型資料集來說，這是一個很好的起點，您可能不想啟動資料庫伺服器。

在此處了解更多資訊：https://docs.docarray.org/user_guide/storing/index_in_memory/

from docarray.index import InMemoryExactNNIndex

# initialize the index
db = InMemoryExactNNIndex[MyDoc]()
# index data
db.index(
    [
        MyDoc(
            title=f"My document {i}",
            title_embedding=embeddings.embed_query(f"query {i}"),
            year=i,
            color=random.choice(["red", "green", "blue"]),
        )
        for i in range(100)
    ]
)
# optionally, you can create a filter query
filter_query = {"year": {"$lte": 90}}
# create a retriever
retriever = DocArrayRetriever(
    index=db,
    embeddings=embeddings,
    search_field="title_embedding",
    content_field="title",
    filters=filter_query,
)

# find the relevant document
doc = retriever.invoke("some query")
print(doc)
[Document(page_content='My document 56', metadata={'id': '1f33e58b6468ab722f3786b96b20afe6', 'year': 56, 'color': 'red'})]

HnswDocumentIndex​

HnswDocumentIndex 是一種輕量級文件索引實作，可在本機完全執行，最適合小型到中型資料集。它將向量儲存在磁碟中的 hnswlib 中，並將所有其他資料儲存在 SQLite 中。

在此處了解更多資訊：https://docs.docarray.org/user_guide/storing/index_hnswlib/

from docarray.index import HnswDocumentIndex

# initialize the index
db = HnswDocumentIndex[MyDoc](work_dir="hnsw_index")

# index data
db.index(
    [
        MyDoc(
            title=f"My document {i}",
            title_embedding=embeddings.embed_query(f"query {i}"),
            year=i,
            color=random.choice(["red", "green", "blue"]),
        )
        for i in range(100)
    ]
)
# optionally, you can create a filter query
filter_query = {"year": {"$lte": 90}}
# create a retriever
retriever = DocArrayRetriever(
    index=db,
    embeddings=embeddings,
    search_field="title_embedding",
    content_field="title",
    filters=filter_query,
)

# find the relevant document
doc = retriever.invoke("some query")
print(doc)
[Document(page_content='My document 28', metadata={'id': 'ca9f3f4268eec7c97a7d6e77f541cb82', 'year': 28, 'color': 'red'})]

WeaviateDocumentIndex​

WeaviateDocumentIndex 是一個文件索引，建立在 Weaviate 向量資料庫之上。

在此處了解更多資訊：https://docs.docarray.org/user_guide/storing/index_weaviate/

# There's a small difference with the Weaviate backend compared to the others.
# Here, you need to 'mark' the field used for vector search with 'is_embedding=True'.
# So, let's create a new schema for Weaviate that takes care of this requirement.

from pydantic import Field


class WeaviateDoc(BaseDoc):
    title: str
    title_embedding: NdArray[32] = Field(is_embedding=True)
    year: int
    color: str
from docarray.index import WeaviateDocumentIndex

# initialize the index
dbconfig = WeaviateDocumentIndex.DBConfig(host="https://127.0.0.1:8080")
db = WeaviateDocumentIndex[WeaviateDoc](db_config=dbconfig)

# index data
db.index(
    [
        MyDoc(
            title=f"My document {i}",
            title_embedding=embeddings.embed_query(f"query {i}"),
            year=i,
            color=random.choice(["red", "green", "blue"]),
        )
        for i in range(100)
    ]
)
# optionally, you can create a filter query
filter_query = {"path": ["year"], "operator": "LessThanEqual", "valueInt": "90"}
# create a retriever
retriever = DocArrayRetriever(
    index=db,
    embeddings=embeddings,
    search_field="title_embedding",
    content_field="title",
    filters=filter_query,
)

# find the relevant document
doc = retriever.invoke("some query")
print(doc)
[Document(page_content='My document 17', metadata={'id': '3a5b76e85f0d0a01785dc8f9d965ce40', 'year': 17, 'color': 'red'})]

ElasticDocIndex​

ElasticDocIndex 是一個文件索引，建立在 ElasticSearch 之上

在此處了解更多資訊

from docarray.index import ElasticDocIndex

# initialize the index
db = ElasticDocIndex[MyDoc](
    hosts="https://127.0.0.1:9200", index_name="docarray_retriever"
)

# index data
db.index(
    [
        MyDoc(
            title=f"My document {i}",
            title_embedding=embeddings.embed_query(f"query {i}"),
            year=i,
            color=random.choice(["red", "green", "blue"]),
        )
        for i in range(100)
    ]
)
# optionally, you can create a filter query
filter_query = {"range": {"year": {"lte": 90}}}
# create a retriever
retriever = DocArrayRetriever(
    index=db,
    embeddings=embeddings,
    search_field="title_embedding",
    content_field="title",
    filters=filter_query,
)

# find the relevant document
doc = retriever.invoke("some query")
print(doc)
[Document(page_content='My document 46', metadata={'id': 'edbc721bac1c2ad323414ad1301528a4', 'year': 46, 'color': 'green'})]

QdrantDocumentIndex​

QdrantDocumentIndex 是一個文件索引，建立在 Qdrant 向量資料庫之上

在此處了解更多資訊

from docarray.index import QdrantDocumentIndex
from qdrant_client.http import models as rest

# initialize the index
qdrant_config = QdrantDocumentIndex.DBConfig(path=":memory:")
db = QdrantDocumentIndex[MyDoc](qdrant_config)

# index data
db.index(
    [
        MyDoc(
            title=f"My document {i}",
            title_embedding=embeddings.embed_query(f"query {i}"),
            year=i,
            color=random.choice(["red", "green", "blue"]),
        )
        for i in range(100)
    ]
)
# optionally, you can create a filter query
filter_query = rest.Filter(
    must=[
        rest.FieldCondition(
            key="year",
            range=rest.Range(
                gte=10,
                lt=90,
            ),
        )
    ]
)
WARNING:root:Payload indexes have no effect in the local Qdrant. Please use server Qdrant if you need payload indexes.
# create a retriever
retriever = DocArrayRetriever(
    index=db,
    embeddings=embeddings,
    search_field="title_embedding",
    content_field="title",
    filters=filter_query,
)

# find the relevant document
doc = retriever.invoke("some query")
print(doc)
[Document(page_content='My document 80', metadata={'id': '97465f98d0810f1f330e4ecc29b13d20', 'year': 80, 'color': 'blue'})]

使用 HnswDocumentIndex 的電影檢索​

movies = [
    {
        "title": "Inception",
        "description": "A thief who steals corporate secrets through the use of dream-sharing technology is given the task of planting an idea into the mind of a CEO.",
        "director": "Christopher Nolan",
        "rating": 8.8,
    },
    {
        "title": "The Dark Knight",
        "description": "When the menace known as the Joker wreaks havoc and chaos on the people of Gotham, Batman must accept one of the greatest psychological and physical tests of his ability to fight injustice.",
        "director": "Christopher Nolan",
        "rating": 9.0,
    },
    {
        "title": "Interstellar",
        "description": "Interstellar explores the boundaries of human exploration as a group of astronauts venture through a wormhole in space. In their quest to ensure the survival of humanity, they confront the vastness of space-time and grapple with love and sacrifice.",
        "director": "Christopher Nolan",
        "rating": 8.6,
    },
    {
        "title": "Pulp Fiction",
        "description": "The lives of two mob hitmen, a boxer, a gangster's wife, and a pair of diner bandits intertwine in four tales of violence and redemption.",
        "director": "Quentin Tarantino",
        "rating": 8.9,
    },
    {
        "title": "Reservoir Dogs",
        "description": "When a simple jewelry heist goes horribly wrong, the surviving criminals begin to suspect that one of them is a police informant.",
        "director": "Quentin Tarantino",
        "rating": 8.3,
    },
    {
        "title": "The Godfather",
        "description": "An aging patriarch of an organized crime dynasty transfers control of his empire to his reluctant son.",
        "director": "Francis Ford Coppola",
        "rating": 9.2,
    },
]
import getpass
import os

if "OPENAI_API_KEY" not in os.environ:
    os.environ["OPENAI_API_KEY"] = getpass.getpass("OpenAI API Key:")
OpenAI API Key: ········
from docarray import BaseDoc, DocList
from docarray.typing import NdArray
from langchain_openai import OpenAIEmbeddings


# define schema for your movie documents
class MyDoc(BaseDoc):
    title: str
    description: str
    description_embedding: NdArray[1536]
    rating: float
    director: str


embeddings = OpenAIEmbeddings()


# get "description" embeddings, and create documents
docs = DocList[MyDoc](
    [
        MyDoc(
            description_embedding=embeddings.embed_query(movie["description"]), **movie
        )
        for movie in movies
    ]
)
API 參考：OpenAIEmbeddings
from docarray.index import HnswDocumentIndex

# initialize the index
db = HnswDocumentIndex[MyDoc](work_dir="movie_search")

# add data
db.index(docs)

一般檢索器​

from langchain_community.retrievers import DocArrayRetriever

# create a retriever
retriever = DocArrayRetriever(
    index=db,
    embeddings=embeddings,
    search_field="description_embedding",
    content_field="description",
)

# find the relevant document
doc = retriever.invoke("movie about dreams")
print(doc)
API 參考：DocArrayRetriever
[Document(page_content='A thief who steals corporate secrets through the use of dream-sharing technology is given the task of planting an idea into the mind of a CEO.', metadata={'id': 'f1649d5b6776db04fec9a116bbb6bbe5', 'title': 'Inception', 'rating': 8.8, 'director': 'Christopher Nolan'})]

帶有過濾器的檢索器​

from langchain_community.retrievers import DocArrayRetriever

# create a retriever
retriever = DocArrayRetriever(
    index=db,
    embeddings=embeddings,
    search_field="description_embedding",
    content_field="description",
    filters={"director": {"$eq": "Christopher Nolan"}},
    top_k=2,
)

# find relevant documents
docs = retriever.invoke("space travel")
print(docs)
API 參考：DocArrayRetriever
[Document(page_content='Interstellar explores the boundaries of human exploration as a group of astronauts venture through a wormhole in space. In their quest to ensure the survival of humanity, they confront the vastness of space-time and grapple with love and sacrifice.', metadata={'id': 'ab704cc7ae8573dc617f9a5e25df022a', 'title': 'Interstellar', 'rating': 8.6, 'director': 'Christopher Nolan'}), Document(page_content='A thief who steals corporate secrets through the use of dream-sharing technology is given the task of planting an idea into the mind of a CEO.', metadata={'id': 'f1649d5b6776db04fec9a116bbb6bbe5', 'title': 'Inception', 'rating': 8.8, 'director': 'Christopher Nolan'})]
from langchain_community.retrievers import DocArrayRetriever

# create a retriever
retriever = DocArrayRetriever(
    index=db,
    embeddings=embeddings,
    search_field="description_embedding",
    content_field="description",
    filters={"rating": {"$gte": 8.7}},
    search_type="mmr",
    top_k=3,
)

# find relevant documents
docs = retriever.invoke("action movies")
print(docs)
API 參考：DocArrayRetriever
[Document(page_content="The lives of two mob hitmen, a boxer, a gangster's wife, and a pair of diner bandits intertwine in four tales of violence and redemption.", metadata={'id': 'e6aa313bbde514e23fbc80ab34511afd', 'title': 'Pulp Fiction', 'rating': 8.9, 'director': 'Quentin Tarantino'}), Document(page_content='A thief who steals corporate secrets through the use of dream-sharing technology is given the task of planting an idea into the mind of a CEO.', metadata={'id': 'f1649d5b6776db04fec9a116bbb6bbe5', 'title': 'Inception', 'rating': 8.8, 'director': 'Christopher Nolan'}), Document(page_content='When the menace known as the Joker wreaks havoc and chaos on the people of Gotham, Batman must accept one of the greatest psychological and physical tests of his ability to fight injustice.', metadata={'id': '91dec17d4272041b669fd113333a65f7', 'title': 'The Dark Knight', 'rating': 9.0, 'director': 'Christopher Nolan'})]

此頁面是否對您有幫助？