xiaoming-01ai / descartes

descartes

Introduction

Nearest neighbor search algorithm descartes, based on fng, providing Python API.

Install & Usage

1. Installation

1.1 Configure environment dependencies (recommended in the Docker container created by the Ubuntu image)

sudo apt update && sudo apt install -y git cmake g++ python3 python3-setuptools python3-pip libblas-dev liblapack-dev '''

1.2 Download code

git clone https://github.com/xiaoming-01ai/descartes.git

1.3 Configuration & installation

pip3 install descartes/descartes-*-linux_x86_64.whl

2. Usage

python3 descartes.py

The following describes the algorithm execution logic of descartes.py in steps

2.1 Data load

Method 1: Randomly generate data

n, d = 10000, 128
X = np.random.randn(n, d)

Method 2: Load data from dataset (sift_base.fvecs as an example)

def ivecs_read(fname):
    a = np.fromfile(fname, dtype='int32')
    d = a[0]
    return a.reshape(-1, d + 1)[:, 1:].copy()
def fvecs_read(fname):
    return ivecs_read(fname).view('float32')

X = fvecs_read("sift_base.fvecs")

2.2 Build Index

index = py01ai.FNGIndex(metric="square_l2", dimension=dimension))
index.init(count=1000000, m=32, s=1, l=200)
index.add_vector(input=X)

2.3 Search

v = np.random.randn(1, 128)
results = index.search(query=v, topk=10)
print(results)

3. Performance

glove-25-angular

glove-50-angular

glove-100-angular

glove-200-angular

sift-128-euclidean

fashion-mnist-784-euclidean

nytimes-256-angular

gist-960-euclidean

4. Team members

The team members come from 01AI: Xiaoming Peng, Gaofei Wang