Basic Usage

You can use fast_hdbscan as a drop in replacement for hdbscan for any basic usage of hdbscan on data with Euclidean distances. That is, fast_hdbscan uses the exact same interface as hdbscan and a subset of the parameters – so whenever you are using hdbscan with the only parameters that fast_hdbscan supports then you can simply swap them. Let’s load libraries and grab some data to demonstrate:

import numpy as np
import fast_hdbscan

import requests
from io import BytesIO
import seaborn as sns

sns.set(rc={"figure.figsize":(8,8)})
data_request = requests.get(
    "https://github.com/scikit-learn-contrib/hdbscan/blob/master/notebooks/clusterable_data.npy?raw=true"
)
data = np.load(BytesIO(data_request.content))

OMP: Info #276: omp_set_nested routine deprecated, please use omp_set_max_active_levels instead.

Before we do any clustering lets plot the data so we can see what we are trying to cluster:

sns.scatterplot(x=data.T[0], y=data.T[1], alpha=0.5)

<AxesSubplot: >

To use fast_hdbscan you can apply it exactly as you would hdbscan. In this case, following the approach taken by hdbscan in its documentation, we’ll use a min_cluster_size of 15 and call fit_predict on the data we wish to cluster.

cluster_labels = fast_hdbscan.HDBSCAN(min_cluster_size=15).fit_predict(data)

We can then plot the results to see what we are, in fact, performing a standard HDBSCAN clustering – just with full multicore support.

sns.scatterplot(
    x=data.T[0],
    y=data.T[1],
    alpha=0.5,
    hue=cluster_labels,
    style=cluster_labels<0,
    size=cluster_labels<0,
    palette="tab10",
    legend=False
)

<AxesSubplot: >

And that’s all there is to it in terms of getting started with fast_hdbscan. Do note, however, that if you want to use hdbscan features such as support for precomputed distances matrics, metrics other than Euclidean, prediction or soft clustering etc. then you’ll need to fall back to the hdbscan library. The goal of this library is to run on the common case (large volumes of low dimensional Euclidean data) as fast as possible with full multicore support.