mirror of https://github.com/hwchase17/langchain synced 2024-11-08 07:10:35 +00:00

Integrate Clickhouse as Vector Store (#5650 )

<!--
Thank you for contributing to LangChain! Your PR will appear in our
release under the title you set. Please make sure it highlights your
valuable contribution.

Replace this with a description of the change, the issue it fixes (if
applicable), and relevant context. List any dependencies required for
this change.

After you're done, someone will review your PR. They may suggest
improvements. If no one reviews your PR within a few days, feel free to
@-mention the same people again, as notifications can get lost.

Finally, we'd love to show appreciation for your contribution - if you'd
like us to shout you out on Twitter, please also include your handle!
-->

#### Description

This PR is mainly to integrate open source version of ClickHouse as
Vector Store as it is easy for both local development and adoption of
LangChain for enterprises who already have large scale clickhouse
deployment.

ClickHouse is a open source real-time OLAP database with full SQL
support and a wide range of functions to assist users in writing
analytical queries. Some of these functions and data structures perform
distance operations between vectors, [enabling ClickHouse to be used as
a vector
database](https://clickhouse.com/blog/vector-search-clickhouse-p1).
Recently added ClickHouse capabilities like [Approximate Nearest
Neighbour (ANN)
indices](https://clickhouse.com/docs/en/engines/table-engines/mergetree-family/annindexes)
support faster approximate matching of vectors and provide a promising
development aimed to further enhance the vector matching capabilities of
ClickHouse.

In LangChain, some ClickHouse based commercial variant vector stores
like
[Chroma](https://github.com/hwchase17/langchain/blob/master/langchain/vectorstores/chroma.py)
and
[MyScale](https://github.com/hwchase17/langchain/blob/master/langchain/vectorstores/myscale.py),
etc are already integrated, but for some enterprises with large scale
Clickhouse clusters deployment, it will be more straightforward to
upgrade existing clickhouse infra instead of moving to another similar
vector store solution, so we believe it's a valid requirement to
integrate open source version of ClickHouse as vector store.

As `clickhouse-connect` is already included by other integrations, this
PR won't include any new dependencies.

#### Before submitting

<!-- If you're adding a new integration, please include:

1. Added a test for the integration:
https://github.com/haoch/langchain/blob/clickhouse/tests/integration_tests/vectorstores/test_clickhouse.py
2. Added an example notebook and document showing its use: 
* Notebook:
https://github.com/haoch/langchain/blob/clickhouse/docs/modules/indexes/vectorstores/examples/clickhouse.ipynb
* Doc:
https://github.com/haoch/langchain/blob/clickhouse/docs/integrations/clickhouse.md

See contribution guidelines for more information on how to write tests,
lint
etc:


https://github.com/hwchase17/langchain/blob/master/.github/CONTRIBUTING.md
-->

1. Added a test for the integration:
https://github.com/haoch/langchain/blob/clickhouse/tests/integration_tests/vectorstores/test_clickhouse.py
2. Added an example notebook and document showing its use: 
* Notebook:
https://github.com/haoch/langchain/blob/clickhouse/docs/modules/indexes/vectorstores/examples/clickhouse.ipynb
* Doc:
https://github.com/haoch/langchain/blob/clickhouse/docs/integrations/clickhouse.md


#### Who can review?

Tag maintainers/contributors who might be interested:

<!-- For a quicker response, figure out the right person to tag with @

  @hwchase17 - project lead

  Tracing / Callbacks
  - @agola11

  Async
  - @agola11

  DataLoaders
  - @eyurtsev

  Models
  - @hwchase17
  - @agola11

  Agents / Tools / Toolkits
  - @vowelparrot

  VectorStores / Retrievers / Memory
  - @dev2049

 -->
 
@hwchase17 @dev2049 Could you please help review?

---------

Co-authored-by: Dev 2049 <dev.dev2049@gmail.com>

2023-06-05 13:32:04 -07:00

2.7 KiB

Raw Blame History

ClickHouse

This page covers how to use ClickHouse Vector Search within LangChain.

ClickHouse is a open source real-time OLAP database with full SQL support and a wide range of functions to assist users in writing analytical queries. Some of these functions and data structures perform distance operations between vectors, enabling ClickHouse to be used as a vector database.

Due to the fully parallelized query pipeline, ClickHouse can process vector search operations very quickly, especially when performing exact matching through a linear scan over all rows, delivering processing speed comparable to dedicated vector databases.

High compression levels, tunable through custom compression codecs, enable very large datasets to be stored and queried. ClickHouse is not memory-bound, allowing multi-TB datasets containing embeddings to be queried.

The capabilities for computing the distance between two vectors are just another SQL function and can be effectively combined with more traditional SQL filtering and aggregation capabilities. This allows vectors to be stored and queried alongside metadata, and even rich text, enabling a broad array of use cases and applications.

Finally, experimental ClickHouse capabilities like Approximate Nearest Neighbour (ANN) indices support faster approximate matching of vectors and provide a promising development aimed to further enhance the vector matching capabilities of ClickHouse.

Installation

Install clickhouse server by binary or docker image
Install the Python SDK with pip install clickhouse-connect

Configure clickhouse vector index

Customize ClickhouseSettings object with parameters

```python
from langchain.vectorstores import ClickHouse, ClickhouseSettings
config = ClickhouseSettings(host="<clickhouse-server-host>", port=8123, ...)
index = Clickhouse(embedding_function, config)
index.add_documents(...)
```

Wrappers

supported functions:

add_texts
add_documents
from_texts
from_documents
similarity_search
asimilarity_search
similarity_search_by_vector
asimilarity_search_by_vector
similarity_search_with_relevance_scores

VectorStore

There exists a wrapper around open source Clickhouse database, allowing you to use it as a vectorstore, whether for semantic search or similar example retrieval.

To import this vectorstore:

from langchain.vectorstores import Clickhouse

For a more detailed walkthrough of the MyScale wrapper, see this notebook

2.7 KiB Raw Blame History