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
- 💾 CRUD 操作: 完整的数据增删改查功能(upsert, update, delete, truncate, overwrite)
- 🛡️ 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
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. CRUD 操作
支持完整的数据增删改查操作:
# Upsert(插入或更新)
# 单条 upsert
table.upsert_one(
index_column="id",
values=[1, "更新的内容", [0.1, 0.2, 0.3]],
column_names=["id", "content", "vector"],
update=True # 冲突时执行更新
)
# 批量 upsert
table.upsert_multi(
index_column="id",
values=[
[1, "内容1", [0.1, 0.2, 0.3]],
[2, "内容2", [0.4, 0.5, 0.6]]
],
column_names=["id", "content", "vector"],
update=True,
update_columns=["content", "vector"] # 指定更新的列
)
# Update(更新)
table.update(
columns=["content", "updated_at"],
values=["新内容", "2026-01-22"],
condition="id = 1"
)
# 带表别名和 FROM 子句的更新
table.update(
columns=["status"],
values=["active"],
table_alias="t1",
from_table="other_table",
from_alias="t2",
condition="t1.id = t2.ref_id AND t2.flag = true"
)
# Delete(删除)
table.delete("id > 100") # 删除满足条件的记录
# Truncate(清空)
table.truncate() # 清空表中所有数据,保留表结构
# Overwrite(覆盖)
# 方式1:使用值覆盖
table.overwrite(
values=[
[1, "新数据1", [0.1, 0.2, 0.3]],
[2, "新数据2", [0.4, 0.5, 0.6]]
]
)
# 方式2:使用查询结果覆盖
source_table = client.open_table("source_table")
table.overwrite(
values_expression=source_table.select("*").where("active = true")
)
# Drop(删除表)
table.drop() # 删除整个表
5. 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()
6. 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()
7. 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()
8. 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
# }
# }
# }
9. 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")
10. 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()
11. 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()
)
12. 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tokenstemmer: 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tokenremovepunct: 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upsert_one(): 插入或更新单条记录(ON CONFLICT)upsert_multi(): 批量插入或更新记录update(): 更新指定条件的记录delete(): 删除满足条件的记录truncate(): 清空表中所有数据overwrite(): 覆盖表中所有数据drop(): 删除表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
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file holo_search_sdk-0.4.1.tar.gz.
File metadata
- Download URL: holo_search_sdk-0.4.1.tar.gz
- Upload date:
- Size: 56.7 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.1.0 CPython/3.8.6
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
9aea98b6078b9202abb568ed69d798d5e0505d2b4cc3a136a6aa84402bcd2133
|
|
| MD5 |
1e32174b5de0b0c127b9fef2c1b70783
|
|
| BLAKE2b-256 |
0bb870a4999dabbba15e98d201a7399aab76ab96931ad1a27392ba5252cc9165
|
File details
Details for the file holo_search_sdk-0.4.1-py3-none-any.whl.
File metadata
- Download URL: holo_search_sdk-0.4.1-py3-none-any.whl
- Upload date:
- Size: 32.6 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.1.0 CPython/3.8.6
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
ef1059895ea936ff6a087f68dac92bd1ae0320e51ec5b1d4e7bed7a5dd6beb45
|
|
| MD5 |
7e9ecf2ae2cf10ea949ad91a3141c8f2
|
|
| BLAKE2b-256 |
8a303059a979272f90a96f31b167443cc27675e8cc8f970a3ac0cb80bf803c70
|