| Name | affine-vectordb JSON |
| Version |
0.1.0
JSON |
| download |
| home_page | None |
| Summary | A unified interface for vector databases |
| upload_time | 2024-11-09 14:15:50 |
| maintainer | None |
| docs_url | None |
| author | None |
| requires_python | >=3.8 |
| license | None |
| keywords |
|
| VCS |
|
| bugtrack_url |
|
| requirements |
No requirements were recorded.
|
| Travis-CI |
No Travis.
|
| coveralls test coverage |
No coveralls.
|
# affine
Affine is a Python library for providing a uniform and structured interface to various backing vector databases and approximate nearest neighbor libraries. It allows simple dataclass-like objects to describe collections together with a high-level query syntax for doing filtered vector search.
For vector databases, it currently supports:
- qdrant
- weaviate
- pinecone
For local mode, the following approximate nearest neighbor libraries are supported:
- FAISS
- annoy
- pynndescent
- scikit-learn KDTree
- naive/NumPy
Note: this project is very similar to [vectordb-orm](https://github.com/piercefreeman/vectordb-orm), which looks to be no longer maintained.
## Installation
```bash
pip install affine-vectordb
# or `pip install "affine-vectordb[qdrant]"` for qdrant support
# `pip install "affine-vectordb[weaviate]"` for weaviate support
# `pip install "affine-vectordb[pinecone]"` for pinecone support
```
## Basic Usage
```python
from affine import Collection, Vector, Filter, Query
# Define a collection
class MyCollection(Collection):
vec: Vector[3] # declare a 3-dimensional vector
# support for additional fields for filtering
a: int
b: str
db = LocalEngine()
# Insert vectors
db.insert(MyCollection(vec=[0.1, 0.0, -0.5], a=1, b="foo"))
db.insert(MyCollection(vec=[1.3, 2.1, 3.6], a=2, b="bar"))
db.insert(MyCollection(vec=[-0.1, 0.2, 0.3], a=3, b="foo"))
# Query vectors
result: list[MyCollection] = (
db.query(MyCollection)
.filter(MyCollection.b == "foo")
.similarity([2.8, 1.8, -4.5])
.limit(1)
)
```
## Engines
A fundamental notion of _affine_ are `Engine` classes. All such classes conform to the same API for interchangeabillity (with the exception of a few engine-specific restrictions which are be mentioned below). There are two broad types of engines
1. `LocalEngine`: this does nearest neighbor search on the executing machine, and supports a variety of libraries for the backing nearest neighborsearch (these are called the _backend_ of the local engine).
2. Vector database engines: these are engines that connect to a vector database service, such as QDrant, Weaviate, or Pinecone.
### Vector Databases
The currently supported vector databases are:
| Database | Class | Constructor arguments | Notes |
| -------- | ------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------- |
| Qdrant | `affine.engine.QdrantEngine` | `host: str` hostname to use<br><br>`port: int` port to use | - |
| Weaviate | `affine.engine.WeaviateEngine` | `host: str` hostname to use<br><br>`port: int` port to use | - |
| Pinecone | `affine.engine.PineconeEngine` | `api_key: Union[str, None]` pinecone API key. if not provided, it will be read from the environment variable PINECONE_API_KEY.<br><br>`spec: Union[ServerlessSpec, PodSpec, None]` the PodSpec or ServerlessSpec object. If not provided, a`ServerlessSpec` will be created from the environment variables PINECONE_CLOUD and PINECONE_REGION. | the Pinecone engine has the restriction that every collection must contain exactly one vector attribute. |
### Approximate Nearest Neighbor Libraries
The `LocalEngine` class provides an interface for doing nearest neighbor search on the executing machine, supporting a variety of libraries for the backing nearest neighborsearch. Which one is specified by the `backend` argument to the constructor. For example, to use `annoy`:
```python
from affine.engine.local import LocalEngine, AnnoyBackend
db = LocalEngine(backend=AnnoyBackend(n_tress=10))
```
The options and settings for the various supported backends are as follows:
| Library | Class | Constructor arguments | Notes |
| ------------------- | ---------------------------------------- | ------------------------------------------------------------------------ | ----- |
| naive/numpy | `affine.engine.local.NumPyBackend` | - | - |
| scikit-learn KDTree | `affine.engine.local.KDTreeBackend` | keyword arguments that get passed directly to `sklearn.neighbors.KDTree` | - |
| annoy | `affine.engine.local.AnnoyBackend` | `n_trees: int` number of trees to use<br>`n_jobs: int` defaults to -1 | - |
| FAISS | `affine.engine.local.FAISSBackend` | `index_factory_str: str` | - |
| PyNNDescent | `affine.engine.local.PyNNDescentBackend` | keyword arguments that get passed directly to `pynndescent.NNDescent` | - |
Raw data
{
"_id": null,
"home_page": null,
"name": "affine-vectordb",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.8",
"maintainer_email": null,
"keywords": null,
"author": null,
"author_email": null,
"download_url": "https://files.pythonhosted.org/packages/5b/69/59dcf805e7bcff6ea0f4c2185f43932133e245ae505f974307a8301af923/affine_vectordb-0.1.0.tar.gz",
"platform": null,
"description": "# affine\n\n\n\nAffine is a Python library for providing a uniform and structured interface to various backing vector databases and approximate nearest neighbor libraries. It allows simple dataclass-like objects to describe collections together with a high-level query syntax for doing filtered vector search.\n\nFor vector databases, it currently supports:\n\n- qdrant\n- weaviate\n- pinecone\n\nFor local mode, the following approximate nearest neighbor libraries are supported:\n\n- FAISS\n- annoy\n- pynndescent\n- scikit-learn KDTree\n- naive/NumPy\n\nNote: this project is very similar to [vectordb-orm](https://github.com/piercefreeman/vectordb-orm), which looks to be no longer maintained.\n\n## Installation\n\n```bash\npip install affine-vectordb\n# or `pip install \"affine-vectordb[qdrant]\"` for qdrant support\n# `pip install \"affine-vectordb[weaviate]\"` for weaviate support\n# `pip install \"affine-vectordb[pinecone]\"` for pinecone support\n```\n\n## Basic Usage\n\n```python\nfrom affine import Collection, Vector, Filter, Query\n\n# Define a collection\nclass MyCollection(Collection):\n vec: Vector[3] # declare a 3-dimensional vector\n\n # support for additional fields for filtering\n a: int\n b: str\n\ndb = LocalEngine()\n\n# Insert vectors\ndb.insert(MyCollection(vec=[0.1, 0.0, -0.5], a=1, b=\"foo\"))\ndb.insert(MyCollection(vec=[1.3, 2.1, 3.6], a=2, b=\"bar\"))\ndb.insert(MyCollection(vec=[-0.1, 0.2, 0.3], a=3, b=\"foo\"))\n\n# Query vectors\nresult: list[MyCollection] = (\n db.query(MyCollection)\n .filter(MyCollection.b == \"foo\")\n .similarity([2.8, 1.8, -4.5])\n .limit(1)\n)\n```\n\n## Engines\n\nA fundamental notion of _affine_ are `Engine` classes. All such classes conform to the same API for interchangeabillity (with the exception of a few engine-specific restrictions which are be mentioned below). There are two broad types of engines\n\n1. `LocalEngine`: this does nearest neighbor search on the executing machine, and supports a variety of libraries for the backing nearest neighborsearch (these are called the _backend_ of the local engine).\n\n2. Vector database engines: these are engines that connect to a vector database service, such as QDrant, Weaviate, or Pinecone.\n\n### Vector Databases\n\nThe currently supported vector databases are:\n\n| Database | Class | Constructor arguments | Notes |\n| -------- | ------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------- |\n| Qdrant | `affine.engine.QdrantEngine` | `host: str` hostname to use<br><br>`port: int` port to use | - |\n| Weaviate | `affine.engine.WeaviateEngine` | `host: str` hostname to use<br><br>`port: int` port to use | - |\n| Pinecone | `affine.engine.PineconeEngine` | `api_key: Union[str, None]` pinecone API key. if not provided, it will be read from the environment variable PINECONE_API_KEY.<br><br>`spec: Union[ServerlessSpec, PodSpec, None]` the PodSpec or ServerlessSpec object. If not provided, a`ServerlessSpec` will be created from the environment variables PINECONE_CLOUD and PINECONE_REGION. | the Pinecone engine has the restriction that every collection must contain exactly one vector attribute. |\n\n### Approximate Nearest Neighbor Libraries\n\nThe `LocalEngine` class provides an interface for doing nearest neighbor search on the executing machine, supporting a variety of libraries for the backing nearest neighborsearch. Which one is specified by the `backend` argument to the constructor. For example, to use `annoy`:\n\n```python\nfrom affine.engine.local import LocalEngine, AnnoyBackend\n\ndb = LocalEngine(backend=AnnoyBackend(n_tress=10))\n```\n\nThe options and settings for the various supported backends are as follows:\n\n| Library | Class | Constructor arguments | Notes |\n| ------------------- | ---------------------------------------- | ------------------------------------------------------------------------ | ----- |\n| naive/numpy | `affine.engine.local.NumPyBackend` | - | - |\n| scikit-learn KDTree | `affine.engine.local.KDTreeBackend` | keyword arguments that get passed directly to `sklearn.neighbors.KDTree` | - |\n| annoy | `affine.engine.local.AnnoyBackend` | `n_trees: int` number of trees to use<br>`n_jobs: int` defaults to -1 | - |\n| FAISS | `affine.engine.local.FAISSBackend` | `index_factory_str: str` | - |\n| PyNNDescent | `affine.engine.local.PyNNDescentBackend` | keyword arguments that get passed directly to `pynndescent.NNDescent` | - |\n",
"bugtrack_url": null,
"license": null,
"summary": "A unified interface for vector databases",
"version": "0.1.0",
"project_urls": null,
"split_keywords": [],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "f5c955c6def2ec286136690ad9744b4f9b33f609d53df74a1b434179a6e0f2af",
"md5": "53d3b2296d6a2d74919936699f1ff9c3",
"sha256": "3fd5ab649730c790ca83660e640de6827be456fc4522dd397f1feddf38d27afd"
},
"downloads": -1,
"filename": "affine_vectordb-0.1.0-py3-none-any.whl",
"has_sig": false,
"md5_digest": "53d3b2296d6a2d74919936699f1ff9c3",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.8",
"size": 16141,
"upload_time": "2024-11-09T14:15:48",
"upload_time_iso_8601": "2024-11-09T14:15:48.335044Z",
"url": "https://files.pythonhosted.org/packages/f5/c9/55c6def2ec286136690ad9744b4f9b33f609d53df74a1b434179a6e0f2af/affine_vectordb-0.1.0-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "5b6959dcf805e7bcff6ea0f4c2185f43932133e245ae505f974307a8301af923",
"md5": "853c7a3610973744cc42b317abb5ce09",
"sha256": "c99c24e3ba7744c6c8e28452f0fe2fd739b7ced80ab3c37732d979f0239a5862"
},
"downloads": -1,
"filename": "affine_vectordb-0.1.0.tar.gz",
"has_sig": false,
"md5_digest": "853c7a3610973744cc42b317abb5ce09",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.8",
"size": 20930,
"upload_time": "2024-11-09T14:15:50",
"upload_time_iso_8601": "2024-11-09T14:15:50.206892Z",
"url": "https://files.pythonhosted.org/packages/5b/69/59dcf805e7bcff6ea0f4c2185f43932133e245ae505f974307a8301af923/affine_vectordb-0.1.0.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-11-09 14:15:50",
"github": false,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"lcname": "affine-vectordb"
}
