2023年3月28日

Zilliz 与 OpenAI 入门指南

在Github中打开以Markdown格式查看

寻找你的下一本书

在本笔记本中，我们将介绍如何使用OpenAI生成书籍描述的嵌入向量，并在Zilliz中利用这些嵌入向量来查找相关书籍。本示例的数据集来源于HuggingFace数据集，包含略超过100万条书名-描述对。

首先下载本笔记本所需的库：

  • openai 用于与OpenAI嵌入服务进行通信
  • pymilvus 用于与 Zilliz 实例进行通信
  • datasets 用于下载数据集
  • tqdm 用于显示进度条
! pip install openai pymilvus datasets tqdm
Looking in indexes: https://pypi.org/simple, https://pypi.ngc.nvidia.com
Requirement already satisfied: openai in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (0.27.2)
Requirement already satisfied: pymilvus in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (2.2.2)
Requirement already satisfied: datasets in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (2.10.1)
Requirement already satisfied: tqdm in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (4.64.1)
Requirement already satisfied: requests>=2.20 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from openai) (2.28.2)
Requirement already satisfied: aiohttp in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from openai) (3.8.4)
Requirement already satisfied: ujson<=5.4.0,>=2.0.0 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from pymilvus) (5.1.0)
Requirement already satisfied: grpcio-tools<=1.48.0,>=1.47.0 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from pymilvus) (1.47.2)
Requirement already satisfied: grpcio<=1.48.0,>=1.47.0 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from pymilvus) (1.47.2)
Requirement already satisfied: mmh3<=3.0.0,>=2.0 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from pymilvus) (3.0.0)
Requirement already satisfied: pandas>=1.2.4 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from pymilvus) (1.5.3)
Requirement already satisfied: numpy>=1.17 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from datasets) (1.23.5)
Requirement already satisfied: xxhash in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from datasets) (3.2.0)
Requirement already satisfied: responses<0.19 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from datasets) (0.18.0)
Requirement already satisfied: dill<0.3.7,>=0.3.0 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from datasets) (0.3.6)
Requirement already satisfied: huggingface-hub<1.0.0,>=0.2.0 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from datasets) (0.12.1)
Requirement already satisfied: pyarrow>=6.0.0 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from datasets) (10.0.1)
Requirement already satisfied: multiprocess in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from datasets) (0.70.14)
Requirement already satisfied: pyyaml>=5.1 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from datasets) (5.4.1)
Requirement already satisfied: fsspec[http]>=2021.11.1 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from datasets) (2023.1.0)
Requirement already satisfied: packaging in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from datasets) (23.0)
Requirement already satisfied: frozenlist>=1.1.1 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from aiohttp->openai) (1.3.3)
Requirement already satisfied: async-timeout<5.0,>=4.0.0a3 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from aiohttp->openai) (4.0.2)
Requirement already satisfied: aiosignal>=1.1.2 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from aiohttp->openai) (1.3.1)
Requirement already satisfied: attrs>=17.3.0 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from aiohttp->openai) (22.2.0)
Requirement already satisfied: charset-normalizer<4.0,>=2.0 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from aiohttp->openai) (3.0.1)
Requirement already satisfied: yarl<2.0,>=1.0 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from aiohttp->openai) (1.8.2)
Requirement already satisfied: multidict<7.0,>=4.5 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from aiohttp->openai) (6.0.4)
Requirement already satisfied: six>=1.5.2 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from grpcio<=1.48.0,>=1.47.0->pymilvus) (1.16.0)
Requirement already satisfied: protobuf<4.0dev,>=3.12.0 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from grpcio-tools<=1.48.0,>=1.47.0->pymilvus) (3.20.1)
Requirement already satisfied: setuptools in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from grpcio-tools<=1.48.0,>=1.47.0->pymilvus) (65.6.3)
Requirement already satisfied: typing-extensions>=3.7.4.3 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from huggingface-hub<1.0.0,>=0.2.0->datasets) (4.5.0)
Requirement already satisfied: filelock in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from huggingface-hub<1.0.0,>=0.2.0->datasets) (3.9.0)
Requirement already satisfied: python-dateutil>=2.8.1 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from pandas>=1.2.4->pymilvus) (2.8.2)
Requirement already satisfied: pytz>=2020.1 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from pandas>=1.2.4->pymilvus) (2022.7.1)
Requirement already satisfied: urllib3<1.27,>=1.21.1 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from requests>=2.20->openai) (1.26.14)
Requirement already satisfied: idna<4,>=2.5 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from requests>=2.20->openai) (3.4)
Requirement already satisfied: certifi>=2017.4.17 in /Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages (from requests>=2.20->openai) (2022.12.7)

要让Zilliz启动并运行，请查看这里。在设置好您的账户和数据库后，继续设置以下值：

  • URI: 您的数据库运行的URI地址
  • 用户: 您的数据库用户名
  • PASSWORD: 您的数据库密码
  • COLLECTION_NAME: 在Zilliz中为集合命名的名称
  • DIMENSION: 嵌入向量的维度
  • OPENAI_ENGINE: 使用哪种嵌入模型
  • openai.api_key: 您的OpenAI账户密钥
  • INDEX_PARAM: 用于集合的索引设置
  • QUERY_PARAM: 要使用的搜索参数
  • BATCH_SIZE: 每次批量嵌入和插入的文本数量
import openai

URI = 'your_uri'
TOKEN = 'your_token' # TOKEN == user:password or api_key
COLLECTION_NAME = 'book_search'
DIMENSION = 1536
OPENAI_ENGINE = 'text-embedding-3-small'
openai.api_key = 'sk-your-key'

INDEX_PARAM = {
    'metric_type':'L2',
    'index_type':"AUTOINDEX",
    'params':{}
}

QUERY_PARAM = {
    "metric_type": "L2",
    "params": {},
}

BATCH_SIZE = 1000

Zilliz

本部分内容涉及Zilliz以及为此用例设置数据库。在Zilliz中，我们需要创建一个集合并为其建立索引。

from pymilvus import connections, utility, FieldSchema, Collection, CollectionSchema, DataType

# Connect to Zilliz Database
connections.connect(uri=URI, token=TOKEN)
# Remove collection if it already exists
if utility.has_collection(COLLECTION_NAME):
    utility.drop_collection(COLLECTION_NAME)
# Create collection which includes the id, title, and embedding.
fields = [
    FieldSchema(name='id', dtype=DataType.INT64, is_primary=True, auto_id=True),
    FieldSchema(name='title', dtype=DataType.VARCHAR, max_length=64000),
    FieldSchema(name='description', dtype=DataType.VARCHAR, max_length=64000),
    FieldSchema(name='embedding', dtype=DataType.FLOAT_VECTOR, dim=DIMENSION)
]
schema = CollectionSchema(fields=fields)
collection = Collection(name=COLLECTION_NAME, schema=schema)
# Create the index on the collection and load it.
collection.create_index(field_name="embedding", index_params=INDEX_PARAM)
collection.load()

数据集

Zilliz启动并运行后，我们就可以开始获取数据了。Hugging Face Datasets是一个包含许多不同用户数据集的中心，在本示例中我们使用Skelebor的书籍数据集。该数据集包含超过100万本书的标题-描述对。我们将嵌入每个描述，并将其与标题一起存储在Zilliz中。

import datasets

# Download the dataset and only use the `train` portion (file is around 800Mb)
dataset = datasets.load_dataset('Skelebor/book_titles_and_descriptions_en_clean', split='train')
/Users/filiphaltmayer/miniconda3/envs/haystack/lib/python3.9/site-packages/tqdm/auto.py:22: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html
  from .autonotebook import tqdm as notebook_tqdm
Found cached dataset parquet (/Users/filiphaltmayer/.cache/huggingface/datasets/Skelebor___parquet/Skelebor--book_titles_and_descriptions_en_clean-3596935b1d8a7747/0.0.0/2a3b91fbd88a2c90d1dbbb32b460cf621d31bd5b05b934492fdef7d8d6f236ec)

插入数据

现在我们已经将数据下载到本地机器上，可以开始进行向量化处理并将其插入Zilliz数据库。嵌入函数接收文本输入，并以列表格式返回向量化结果。

# Simple function that converts the texts to embeddings
def embed(texts):
    embeddings = openai.Embedding.create(
        input=texts,
        engine=OPENAI_ENGINE
    )
    return [x['embedding'] for x in embeddings['data']]

下一步将执行实际的插入操作。由于数据点数量庞大，如果你想立即测试，可以提前停止插入单元格并继续。这样做可能会因为数据点减少而降低结果的准确性，但仍应足够好。

from tqdm import tqdm

data = [
    [], # title
    [], # description
]

# Embed and insert in batches
for i in tqdm(range(0, len(dataset))):
    data[0].append(dataset[i]['title'])
    data[1].append(dataset[i]['description'])
    if len(data[0]) % BATCH_SIZE == 0:
        data.append(embed(data[1]))
        collection.insert(data)
        data = [[],[]]

# Embed and insert the remainder 
if len(data[0]) != 0:
    data.append(embed(data[1]))
    collection.insert(data)
    data = [[],[]]

  0%|          | 2999/1032335 [00:19<1:49:30, 156.66it/s]
KeyboardInterrupt

查询数据库

将数据安全插入Zilliz后，我们现在可以执行查询。该查询接收字符串或字符串列表并进行搜索。结果会显示您提供的描述以及包含结果分数、结果标题和结果书籍描述的搜索结果。

import textwrap

def query(queries, top_k = 5):
    if type(queries) != list:
        queries = [queries]
    res = collection.search(embed(queries), anns_field='embedding', param=QUERY_PARAM, limit = top_k, output_fields=['title', 'description'])
    for i, hit in enumerate(res):
        print('Description:', queries[i])
        print('Results:')
        for ii, hits in enumerate(hit):
            print('\t' + 'Rank:', ii + 1, 'Score:', hits.score, 'Title:', hits.entity.get('title'))
            print(textwrap.fill(hits.entity.get('description'), 88))
            print()
query('Book about a k-9 from europe')
Description: Book about a k-9 from europe
Results:
	Rank: 1 Score: 0.3047754764556885 Title: Bark M For Murder
Who let the dogs out? Evildoers beware! Four of mystery fiction's top storytellers are
setting the hounds on your trail -- in an incomparable quartet of crime stories with a
canine edge. Man's (and woman's) best friends take the lead in this phenomenal
collection of tales tense and surprising, humorous and thrilling: New York
Timesbestselling author J.A. Jance's spellbinding saga of a scam-busting septuagenarian
and her two golden retrievers; Anthony Award winner Virginia Lanier's pureblood thriller
featuring bloodhounds and bloody murder; Chassie West's suspenseful stunner about a
life-saving German shepherd and a ghastly forgotten crime; rising star Lee Charles
Kelley's edge-of-your-seat yarn that pits an ex-cop/kennel owner and a yappy toy poodle
against a craven killer.

	Rank: 2 Score: 0.3283390402793884 Title: Texas K-9 Unit Christmas: Holiday Hero\Rescuing Christmas
CHRISTMAS COMES WRAPPED IN DANGER Holiday Hero by Shirlee McCoy Emma Fairchild never
expected to find trouble in sleepy Sagebrush, Texas. But when she's attacked and left
for dead in her own diner, her childhood friend turned K-9 cop Lucas Harwood offers a
chance at justice--and love. Rescuing Christmas by Terri Reed She escaped a kidnapper,
but now a killer has set his sights on K-9 dog trainer Lily Anderson. When fellow
officer Jarrod Evans appoints himself her bodyguard, Lily knows more than her life is at
risk--so is her heart. Texas K-9 Unit: These lawmen solve the toughest cases with the
help of their brave canine partners

	Rank: 3 Score: 0.33899369835853577 Title: Dogs on Duty: Soldiers' Best Friends on the Battlefield and Beyond
When the news of the raid on Osama Bin Laden's compound broke, the SEAL team member that
stole the show was a highly trained canine companion. Throughout history, dogs have been
key contributors to military units. Dorothy Hinshaw Patent follows man's best friend
onto the battlefield, showing readers why dogs are uniquely qualified for the job at
hand, how they are trained, how they contribute to missions, and what happens when they
retire. With full-color photographs throughout and sidebars featuring heroic canines
throughout history, Dogs on Duty provides a fascinating look at these exceptional
soldiers and companions.

	Rank: 4 Score: 0.34207457304000854 Title: Toute Allure: Falling in Love in Rural France
After saying goodbye to life as a successful fashion editor in London, Karen Wheeler is
now happy in her small village house in rural France. Her idyll is complete when she
meets the love of her life - he has shaggy hair, four paws and a wet nose!

	Rank: 5 Score: 0.343595951795578 Title: Otherwise Alone (Evan Arden, #1)
Librarian's note: This is an alternate cover edition for ASIN: B00AP5NNWC. Lieutenant
Evan Arden sits in a shack in the middle of nowhere, waiting for orders that will send
him back home - if he ever gets them. Other than his loyal Great Pyrenees, there's no
one around to break up the monotony. The tedium is excruciating, but it is suddenly
interrupted when a young woman stumbles up his path. "It's only 50-something pages, but
in that short amount of time, the author's awesome writing packs in a whole lotta
character detail. And sets the stage for the series, perfectly." -Maryse.net, 4.5 Stars
He has two choices - pick her off from a distance with his trusty sniper-rifle, or dare
let her approach his cabin and enter his life. Why not? It's been ages, and he is
otherwise alone...