See reference paper: http://www.kdd.org/kdd2016/subtopic/view/skinny-dip-clustering-in-a-sea-of-noise
UniDip is a noise robust clustering algorithm for 1 dimensional numeric data. It recursively extracts peaks of density in the data utilizing the Hartigan Dip-test of Unimodality.
From command line in installation directory:
cd <path to UniDip installation directory>
python setup.py install
Installation via Pip (this will not work with newer versions of Python and Numpy):
pip3.6 install unidip
from unidip import UniDip
# create bi-modal distribution
dat = np.concatenate([np.random.randn(200)-3, np.random.randn(200)+3])
# sort data so returned indices are meaningful
dat = np.msort(dat)
# get start and stop indices of peaks
intervals = UniDip(dat).run()- alpha: control sensitivity as p-value. Default is 0.05. increase to isolate more peaks with less confidence. Or, decrease to isolate only peaks that are least likely to be noise.
- mrg_dst: Defines how close intervals must be before they are merged.
- ntrials: how many trials are run in Hartigan Dip Test more trials adds confidance but takes longer.
intervals = UniDip(dat, alpha=0.001, ntrials=1000, mrg_dst=5).run()