FAISSt DenStream

A performant implementation of the DenStream algorithm that relies heavily on FAISS.

Installation

Without GPU acceleration: pip install faisst-denstream

With GPU acceleration via CUDA 11: pip install faisst-denstream[gpu-cu11]

With GPU acceleration via CUDA 12: pip install faisst-denstream[gpu-cu12]

Basic Usage

import numpy as np

from faisst_denstream.DenStream import DenStream
from random import randint
from sys import stderr
from loguru import logger

logger.remove()
logger.add(stderr, level="INFO")

# Create model
lamb = 0.05
beta = 0.5
mu = 10
epsilon = 2
n_init_points = int(test_dataset_size * 0.25)
stream_speed = 10

model = DenStream(lamb, mu, beta, epsilon, n_init_points, stream_speed)

# Multiple datasets to simulate fitting model to stream
X1 = np.random.normal(loc=randint(0, 10), scale=randint(1, 3), size=(1000, 2))
X2 = np.random.normal(loc=randint(0, 10), scale=randint(1, 3), size=(1000, 2))

# As long as model has consumed at least n_init_points points, `predict` and `fit_predict`
# can be called to get cluster labels for each point
model.fit(X1)
x1_labels = model.predict(X1)

x2_labels = model.fit_predict(X2)