Skip to main content

A Python SDK for database search operations with vector and full-text search capabilities

Project description

Holo Search SDK

一个用于Hologres数据检索操作的 Python SDK,支持向量检索和全文检索功能。

✨ 特性

  • 🔍 向量检索: 基于语义相似性的检索功能
  • 📝 全文检索: 传统的基于关键词的检索
  • 🛡️ 类型安全: 使用类型提示和数据验证
  • 🧩 模块化设计: 清晰的分层架构,便于扩展和维护

📦 安装

从 PyPI 安装

pip install holo-search-sdk

🚀 快速开始

基本使用

import holo_search_sdk as holo

# 连接到数据库
client = holo.connect(
    host="your-host",
    port=80,
    database="your-database",
    access_key_id="your-access-key-id",
    access_key_secret="your-access-key-secret",
    schema="public"
)

# 建立连接
client.connect()

# 打开表
columns = {
    "id": ("INTEGER", "PRIMARY KEY"),
    "content": "TEXT",
    "vector": "FLOAT4[]"
}
table = client.open_table("table_name")

# 插入数据
data = [
    [1, "Hello world", [0.1, 0.2, 0.3]],
    [2, "Python SDK", [0.4, 0.5, 0.6]],
    [3, "Vector search", [0.7, 0.8, 0.9]]
]
table.insert_multi(data, ["id", "content", "vector"])

# 设置向量索引
table.set_vector_index(
    column="vector",
    distance_method="Cosine",
    base_quantization_type="rabitq",
    max_degree=64,
    ef_construction=400
)

# 向量检索
query_vector = [0.1, 0.2, 0.3]
# 限制结果数量
results = table.search_vector(query_vector, "vector").limit(10).fetchall()
# 设置最小距离
results = table.search_vector(query_vector, "vector").min_distance(0.5).fetchall()

# 关闭连接
client.disconnect()

使用上下文管理器

import holo_search_sdk as holo

with holo.connect(
    host="your-host",
    port=80,
    database="your-database",
    access_key_id="your-access-key-id",
    access_key_secret="your-access-key-secret"
) as client:
    client.connect()
    
    # 执行数据库操作
    table = client.open_table("table_name")
    results = table.search_vector([0.1, 0.2, 0.3], "vector_column").fetchall()
    
    # 连接会自动关闭

📚 详细文档

核心概念

1. 客户端 (Client)

客户端是与数据库交互的主要接口:

from holo_search_sdk import connect

# 创建客户端
client = connect(
    host="localhost",
    port=80,
    database="test_db",
    access_key_id="your_key",
    access_key_secret="your_secret"
)

# 建立连接
client.connect()

# 执行 SQL
result = client.execute("SELECT COUNT(*) FROM users", fetch_result=True)

# 表操作
table = client.open_table("table_name")

2. 表操作 (Table Operations)

表是数据存储和搜索的基本单位:

# 打开现有表
table = client.open_table("table_name")

# 检查表是否存在
exists = client.check_table_exist("table_name")

# 删除表
client.drop_table("table_name")

3. 数据插入

支持单条和批量数据插入:

# 插入单条记录
table.insert_one(
    [1, "标题", "内容", [0.1, 0.2, 0.3]],
    ["id", "title", "content", "vector"]
)

# 批量插入
data = [
    [1, "文档1", "内容1", [0.1, 0.2, 0.3]],
    [2, "文档2", "内容2", [0.4, 0.5, 0.6]],
    [3, "文档3", "内容3", [0.7, 0.8, 0.9]]
]
table.insert_multi(data, ["id", "title", "content", "vector"])

4. 向量索引

为向量列创建高效的检索索引:

# 设置单个向量索引
table.set_vector_index(
    column="vector",
    distance_method="Cosine",  # 可选: "Euclidean", "InnerProduct", "Cosine"
    base_quantization_type="rabitq",  # 可选: "sq8", "sq8_uniform", "fp16", "fp32", "rabitq"
    max_degree=64,
    ef_construction=400,
    use_reorder=True,
    precise_quantization_type="fp32",
    max_total_size_to_merge_mb=4096,  # 磁盘合并时数据的最大文件大小,单位MB
    build_thread_count=16  # 索引构建过程中使用的线程数
)

# 设置多个向量索引
table.set_vector_indexes({
    "content_vector": {
        "distance_method": "Cosine",
        "base_quantization_type": "rabitq",
        "max_degree": 64,
        "ef_construction": 400,
        "use_reorder": True,
        "precise_quantization_type": "fp32",
        "max_total_size_to_merge_mb": 4096,
        "build_thread_count": 16
    },
    "title_vector": {
        "distance_method": "Euclidean",
        "base_quantization_type": "rabitq",
        "max_degree": 32,
        "ef_construction": 200,
        "use_reorder": True,
        "precise_quantization_type": "fp32",
        "max_total_size_to_merge_mb": 4096,
        "build_thread_count": 16
    }
})

# 删除所有向量索引
table.delete_vector_indexes()

5. 向量检索

执行语义相似性检索:

# 基本向量检索
query_vector = [0.1, 0.2, 0.3]
results = table.search_vector(
    vector=query_vector,
    column="vector",
    distance_method="Cosine"
).fetchall()

# 带输出别名的检索
results = table.search_vector(
    vector=query_vector,
    column="vector",
    output_name="similarity_score",
    distance_method="Cosine"
).fetchall()

6. 数据查询

支持基于主键的精确查询:

# 根据主键查询单条记录
result = table.get_by_key(
    key_column="id",
    key_value=1,
    return_columns=["id", "content", "vector"]  # 可选,不指定则返回所有列
).fetchone()

# 根据主键列表批量查询
results = table.get_multi_by_keys(
    key_column="id", 
    key_values=[1, 2, 3],
    return_columns=["id", "content"]  # 可选,不指定则返回所有列
).fetchall()

7. 向量索引管理

查询和管理向量索引信息:

# 获取向量索引信息
index_info = table.get_vector_index_info()
if index_info:
    print("当前向量索引配置:", index_info)
else:
    print("未找到向量索引配置")

# 索引信息示例返回格式
# {
#     "vector_column": {
#         "algorithm": "HGraph",
#         "distance_method": "Cosine",
#         "builder_params": {
#             "max_degree": 64,
#             "ef_construction": 400,
#             "base_quantization_type": "rabitq",
#             "use_reorder": true,
#             "precise_quantization_type": "fp32",
#             "precise_io_type": "block_memory_io",
#             "max_total_size_to_merge_mb": 4096,
#             "build_thread_count": 16
#         }
#     }
# }

8. 全文检索索引

为文本列创建全文检索索引:

# 创建全文索引
table.create_text_index(
    index_name="ft_idx_content",
    column="content",
    tokenizer="jieba"  # 可选: "jieba", "ik", "icu", "whitespace", "standard", "simple", "keyword", "ngram", "pinyin"
)

# 设置全文索引(修改现有索引的分词器)
table.set_text_index(
    index_name="ft_idx_content",
    tokenizer="ik"
)

# 删除全文索引
table.drop_text_index(index_name="ft_idx_content")

9. 全文检索

执行全文检索查询:

# 基本全文检索
results = table.search_text(
    column="content",
    expression="机器学习",
    return_all_columns=True
).fetchall()

# 带分数返回的全文检索
results = table.search_text(
    column="content",
    expression="深度学习",
    return_score=True,
    return_score_name="relevance_score"
).select(["id", "title", "content"]).fetchall()

# 使用不同的检索模式
# 关键词模式(默认)
results = table.search_text(
    column="content",
    expression="python programming",
    mode="match",
    operator="AND"  # 要求同时包含所有关键词
).fetchall()

# 短语模式
results = table.search_text(
    column="content",
    expression="machine learning",
    mode="phrase"  # 精确短语匹配
).fetchall()

# 自然语言模式
results = table.search_text(
    column="content",
    expression="+python -java",  # 必须包含python,不能包含java
    mode="natural_language"
).fetchall()

# 术语检索
results = table.search_text(
    column="content",
    expression="python",
    mode="term" # 对expression不做分词或其他处理,直接去索引中精确匹配
).fetchall()

10. 高级查询构建

使用查询构建器进行复杂查询:

# 组合全文检索和过滤条件
results = (
    table.search_text(
        column="content",
        expression="人工智能",
        return_score=True,
        return_score_name="score"
    )
    .where("publish_date > '2023-01-01'")
    .order_by("score", "desc")
    .limit(10)
    .fetchall()
)

# 使用过滤器表达式
from holo_search_sdk import Filter, AndFilter, OrFilter, NotFilter

results = (
    table.select(["id", "title", "content"])
    .where(
        AndFilter(
            Filter("category = 'technology'"),
            Filter("views > 1000")
        )
    )
    .order_by("views", "desc")
    .fetchall()
)

# 使用或过滤器表达式
results = (
    table.select(["id", "title", "content"])
    .where(
        Filter("category = 'technology'") | Filter("views > 1000")
    )
    .order_by("views", "desc")
    .fetchall()
)

# 使用分词功能
results = (
    table.select(["id", "content"])
    .select_tokenize(
        column="content",
        tokenizer="jieba",
        output_name="tokens"
    )
    .limit(5)
    .fetchall()
)

11. 表连接查询

支持多表连接查询:

# 内连接
table1 = client.open_table("articles", table_alias="a")
table2 = client.open_table("authors", table_alias="b")

results = (
    table1.select(["a.id", "a.title", "b.name"])
    .inner_join(table2, "a.author_id = b.id")
    .where("a.publish_date > '2023-01-01'")
    .fetchall()
)

# 左连接
results = (
    table1.select(["a.id", "a.title", "b.name"])
    .left_join(table2, "a.author_id = b.id")
    .fetchall()
)

配置选项

连接配置

from holo_search_sdk.types import ConnectionConfig

config = ConnectionConfig(
    host="your-host.com",
    port=80,
    database="production_db",
    access_key_id="user...",
    access_key_secret="secret...",
    schema="analytics"  # 默认为 "public"
)

向量索引配置

  • distance_method: 距离计算方法

    • "Euclidean": 欧几里得距离
    • "InnerProduct": 内积距离
    • "Cosine": 余弦距离
  • base_quantization_type: 基础量化类型

    • "sq8", "sq8_uniform", "fp16", "fp32", "rabitq"
  • max_degree: 图构建过程中每个顶点尝试连接的最近邻数量 (默认: 64)

  • ef_construction: 图构建过程中的检索深度控制 (默认: 400)

  • use_reorder: 是否使用 HGraph 高精度索引 (默认: False)

  • precise_quantization_type: 精确量化类型 (默认: "fp32")

  • precise_io_type: 精确 IO 类型 (默认: "block_memory_io")

  • max_total_size_to_merge_mb: 磁盘合并时数据的最大文件大小,单位MB (默认: 4096)

  • build_thread_count: 索引构建过程中使用的线程数 (默认: 16)

全文检索配置

  • tokenizer: 分词器类型
  • mode: 全文检索模式
    • match:关键词匹配,默认
    • phrase:短语检索
    • natural_language:自然语言检索
    • term:术语检索
  • operator: 关键词检索操作符 (仅适用于match模式, 默认: "OR")
  • 分词过滤器*:
    • lowercase: 将token中的大写字母转为小写
    • stop: 移除停用词token
    • stemmer: 根据对应语言的语法规则将token转化为其对应的词干
    • length: 移除超过指定长度的token
    • removepunct: 移除只包含标点符号字符的token。
    • pinyin: 拼音Token Filter

🔧 API 参考

主要类

  • Client: 数据库客户端,管理连接和表操作
  • HoloTable: 表操作接口,支持数据插入、向量检索和全文检索
  • QueryBuilder: 查询构建器,支持链式调用构建复杂查询
  • ConnectionConfig: 连接配置数据类

过滤器类

  • Filter: 基础过滤器表达式
  • AndFilter: AND 逻辑过滤器
  • OrFilter: OR 逻辑过滤器
  • NotFilter: NOT 逻辑过滤器
  • TextSearchFilter: 全文检索过滤器

主要函数

连接和表管理:

  • connect(): 创建数据库客户端连接
  • open_table(): 打开现有表
  • check_table_exist(): 检查表是否存在
  • drop_table(): 删除表

数据操作:

  • insert_one(): 插入单条记录
  • insert_multi(): 批量插入记录
  • get_by_key(): 根据主键查询单条记录
  • get_multi_by_keys(): 根据主键列表批量查询

向量检索:

  • set_vector_index(): 设置单个向量索引
  • set_vector_indexes(): 设置多个向量索引
  • delete_vector_indexes(): 删除所有向量索引
  • get_vector_index_info(): 获取向量索引信息
  • search_vector(): 执行向量检索

全文检索:

  • create_text_index(): 创建全文索引
  • set_text_index(): 修改全文索引
  • drop_text_index(): 删除全文索引
  • get_index_properties(): 获取索引属性
  • search_text(): 执行全文检索

查询构建:

  • select(): 选择返回的列
  • where(): 添加过滤条件
  • and_where(): 添加 AND 过滤条件
  • or_where(): 添加 OR 过滤条件
  • order_by(): 排序
  • group_by(): 分组
  • limit(): 限制结果数量
  • offset(): 跳过指定数量的结果
  • join(): 表连接
  • inner_join(): 内连接
  • left_join(): 左连接
  • right_join(): 右连接
  • select_tokenize(): 显示分词效果
  • select_text_search(): 在 SELECT 中进行全文检索
  • where_text_search(): 在 WHERE 中进行全文检索过滤

异常类

  • HoloSearchError: 基础异常类
  • ConnectionError: 连接相关错误
  • QueryError: 查询执行错误
  • SqlError: SQL 生成错误
  • TableError: 表操作错误

📄 许可证

本项目采用 MIT 许可证 - 查看 LICENSE.txt 文件了解详情。


Holo Search SDK - 让Hologres向量和全文检索变得简单高效 🚀

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

holo_search_sdk-0.3.0.tar.gz (50.5 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

holo_search_sdk-0.3.0-py3-none-any.whl (30.1 kB view details)

Uploaded Python 3

File details

Details for the file holo_search_sdk-0.3.0.tar.gz.

File metadata

  • Download URL: holo_search_sdk-0.3.0.tar.gz
  • Upload date:
  • Size: 50.5 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.1.0 CPython/3.8.6

File hashes

Hashes for holo_search_sdk-0.3.0.tar.gz
Algorithm Hash digest
SHA256 7223cf8ca98fc76317e6d1780c4538e1136f850598509945035d8d5cd18f83c4
MD5 e50306384a156268bdf69c3a2986436b
BLAKE2b-256 cad121bb642788bc68c7cd15cb57066b36e1d200fd3cc181ef65178c9d2f3866

See more details on using hashes here.

File details

Details for the file holo_search_sdk-0.3.0-py3-none-any.whl.

File metadata

File hashes

Hashes for holo_search_sdk-0.3.0-py3-none-any.whl
Algorithm Hash digest
SHA256 5bfd768aa8871770d8f18e7221e4f9fff98f38347125bb9006f9af1c0278b932
MD5 6cb82aac6859e6afb2b617e877fa924c
BLAKE2b-256 a650ecb079b83de876f5b946b6ede7f677a2079c4c621f7e66a5db32bd0b6f98

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page