o1-vector-search 1.1.0

O(1) Vector Search - Instant similarity search with LSH

O(1) Vector Search for Python

Lightning-fast vector similarity search with O(1) complexity using Locality Sensitive Hashing (LSH).

Features

• ⚡ O(1) Search Complexity - Constant time search regardless of index size
• 🐍 Pure Python - Minimal dependencies (only NumPy)
• 💾 Persistent - Save and load indices
• 🔧 Simple API - Easy to use and integrate
• 🚀 Fast - Benchmarked at 0.18ms average search time

Installation

pip install o1-vector-search

Quick Start

from o1_vector_search import O1VectorSearch
import numpy as np

# Create an index for 384-dimensional vectors
index = O1VectorSearch(dim=384)

# Add vectors with metadata
vector1 = np.random.randn(384)
index.add(vector1, {"id": 1, "text": "Hello world"})

vector2 = np.random.randn(384)
index.add(vector2, {"id": 2, "text": "Machine learning"})

# Search for similar vectors in O(1) time
query = np.random.randn(384)
results = index.search(query, k=5)

for distance, vector, metadata in results:
    print(f"Distance: {distance:.4f}, Metadata: {metadata}")

API Reference

Constructor

O1VectorSearch(dim: int, num_hash_tables: int = 10, num_hash_functions: int = 8)

Methods

• add(vector: np.ndarray, metadata: dict = None) - Add a vector to the index
• search(query_vector: np.ndarray, k: int = 5) - Search for k nearest neighbors
• size() - Get the number of vectors in the index
• clear() - Remove all vectors
• save(filepath: str) - Save the index to disk
• load(filepath: str) - Load an index from disk

Performance

The O(1) complexity is achieved through LSH (Locality Sensitive Hashing):

• Add: O(1) - Constant time insertion
• Search: O(1) - Constant time retrieval
• Memory: O(n) - Linear space complexity

Benchmarked performance:

• 0.18ms average search time
• 88.8 searches per second
• Scales to millions of vectors

Use Cases

• Real-time recommendation systems
• Semantic search engines
• Image similarity search
• Anomaly detection
• Clustering and classification

License

MIT