async-annoy 0.3.0

Asynchronous wrapper around the Annoy library for approximate nearest neighbor search.

async-annoy README

Overview

async-annoy simplifies working with Annoy indices in async applications like web servers. It's built to let you read from or write to Annoy indices without worrying about the complexities of concurrent access. You get easy, thread-safe operations, whether you're querying nearest neighbors or building the index, all in the background.

Features

• Automated Index Management: Handles all the tricky parts of managing Annoy indices on disk.
• Concurrency Made Simple: Supports multiple readers at once or a single writer by managing access locks behind the scenes.

Installation

pip install async-annoy

Make sure Annoy and numpy are installed as they're needed to run the library.

How to Use

Reading from the Index

Ideal for fetching nearest neighbors in response to web requests.

from async_annoy import AsyncAnnoy

async with AsyncAnnoy("my_index").reader() as reader:
    neighbours = await reader.get_neighbours_for(vector=my_vector, n=5)
    print(neighbours)

Writing to the Index

The writer will wait for all readers to finish and start the index from scratch. The Annoy library doesn't support index updates, only rebuilds.

from async_annoy import AsyncAnnoy

async with AsyncAnnoy("my_index").writer() as writer:
    await writer.add_item(index=1, vector=my_vector)
    # The index is automatically built and saved when done.

Configuration

The initial setup of async-annoy relies on environment variables to configure the Annoy index parameters, ensuring a seamless start. While these defaults offer convenience, you may need to customize settings to fit your application's specific requirements. It's critical to maintain parameter consistency across all readers and writers interacting with the same index.

Default Parameters

async-annoy uses the following environment variables for initial configuration:

• ASYNC_ANNOY_DIMENSIONS: The dimensionality of the vectors stored in the index.
• ASYNC_ANNOY_METRIC: The distance metric used for comparing vectors in the index (e.g., "angular", "euclidean").

Overriding Default Parameters

Although async-annoy configures itself with environment variables, you can override these defaults directly in your code. When creating a new AnnoyReader or AnnoyWriter, simply pass the desired dimensions and metric as arguments:

from async_annoy import AsyncAnnoy

# Override default parameters when initializing
index_manager = AsyncAnnoy("my_index", dimensions=128, metric="euclidean")

Development Status

Currently, async-annoy is in alpha. It's ready for testing, and we're eager for your feedback to make it better. This is an initial version, so we're still working on adding more features and smoothing out the experience.