CREATE TABLE images(`_file` LowCardinality(String),`caption` String,`image_embedding` Array(Float32))ENGINE = MergeTree;

SELECT_file,caption,cosineDistance(-- An embedding of your "input" dog picture[0.5736801028251648, 0.2516217529773712, ...,-0.6825592517852783],image_embedding) AS scoreFROM imagesORDER BY score ASCLIMIT 10

SELECTName,embedding,L2Distance(embedding, [0.1, 0.2, 0.3, 0.4, 0.5]) AS scoreFROM ann_index_exampleORDER BY score ASCLIMIT 5
Query id: 9de95e7f-79e1-4d02-81ff-43f7ace14072
┌─Name──┬─embedding────────────────────────────────────────────────┬───────────────score─┐│ anabd │ [0.109578826,0.17093645,0.1863009,0.4694911,0.56862974]│0.1529803320145931 ││ ewzos │ [0.24622843,0.28853804,0.42360082,0.5199647,0.5086616] │ 0.24282803026289437 ││ avajo │ [0.24545254,0.36559477,0.26052764,0.31315225,0.53613853] │ 0.24286758135453868 ││ wyayw │ [0.032240078,0.18597832,0.31064722,0.21862713,0.6575932] │0.2502660809473027 ││ adxir │ [0.1420014,0.11847292,0.41707036,0.41038868,0.71311146]│0.2600782018429016 │└───────┴──────────────────────────────────────────────────────────┴─────────────────────┘
5 rows in set. Elapsed: 0.006 sec. Processed 1.01 thousand rows, 42.42 KB (178.20 thousand rows/s., 7.48 MB/s.)Peak memory usage: 121.19 KiB.

pip install -qU langchain_community clickhouse-connect

from langchain_community.vectorstores import Clickhouse, ClickhouseSettingsfrom langchain_openai import OpenAIEmbeddings
# 配置 ClickHouse 设置settings = ClickhouseSettings(table="clickhouse_example")
# 实例化 ClickHouse 向量存储vector_store = Clickhouse(embedding=OpenAIEmbeddings(), config=settings)

from langchain_core.documents import Document
# 定义文档document_1 = Document(page_content="foo", metadata={"baz": "bar"})document_2 = Document(page_content="thud", metadata={"bar": "baz"})document_3 = Document(page_content="i will be deleted :(")
# 添加文档到向量存储documents = [document_1, document_2, document_3]ids = ["1", "2", "3"]vector_store.add_documents(documents=documents, ids=ids)

# 删除指定 ID 的文档vector_store.delete(ids=["3"])

# 进行向量搜索results = vector_store.similarity_search(query="thud", k=1)
# 输出搜索结果for doc in results:    print(f"* {doc.page_content} [{doc.metadata}]")

# 使用过滤条件进行向量搜索results = vector_store.similarity_search(query="thud", k=1, filter="metadata.baz='bar'")
# 输出搜索结果for doc in results:    print(f"* {doc.page_content} [{doc.metadata}]")

# 进行带分数的向量搜索results = vector_store.similarity_search_with_score(query="qux", k=1)
# 输出搜索结果和相似度得分for doc, score in results:    print(f"* [SIM={score:3f}] {doc.page_content} [{doc.metadata}]")

# 异步添加文档# await vector_store.aadd_documents(documents=documents, ids=ids)
# 异步删除文档# await vector_store.adelete(ids=["3"])
# 异步向量搜索# results = await vector_store.asimilarity_search(query="thud", k=1)
# 异步带分数的向量搜索results = await vector_store.asimilarity_search_with_score(query="qux", k=1)
# 输出搜索结果和相似度得分for doc, score in results:    print(f"* [SIM={score:3f}] {doc.page_content} [{doc.metadata}]")