Discovering clusters using HDBSCAN

  run this example as a notebook — or grab the .ipynb (also linked at the bottom of the page)

To make use of HDBSCAN as the clustering algorithm used to discover clusters, you must specify it as the cluster argument. If you wish to specify HDBSCAN parameters you will need the dictionary form, which includes both the model name and its keyword arguments (under the ‘kwargs’ key). Since HDBSCAN does not require the number of clusters, n_clusters does not need to be set. Note that HDBSCAN can also label some observations as noise (label -1); those points are colored as their own group. Because the rows of the mushrooms dataset are unordered samples, we plot them as points (‘.’).

/home/docs/checkouts/readthedocs.org/user_builds/hypertools/checkouts/latest/hypertools/plot/backend.py:1353: UserWarning: Failed to switch to any interactive backend (TkAgg, QtAgg, Qt5Agg, Qt4Agg, GTK4Agg, GTK3Agg, WXAgg). Falling back to 'Agg'.
  warnings.warn(BACKEND_WARNING)

# Code source: Andrew Heusser and Leland McInnes
# License: MIT

# import
import hypertools as hyp

# load example data
data = hyp.load('mushrooms')

# plot
hyp.plot(data, '.', cluster={'model': 'HDBSCAN',
                             'kwargs': {'min_samples': 5,
                                        'min_cluster_size': 30}})