Compute Natural Breaks (Fisher-Jenks algorithm)


Keywords
data-classification, jenks-fisher, python-library
License
MIT
Install
pip install jenkspy==0.1.0

Documentation

Fast Fisher-Jenks breaks for Python

Compute "natural breaks" (Fisher-Jenks algorithm) on list / tuple / array / numpy.ndarray of integers/floats.

The algorithm implemented by this library is also sometimes referred to as Fisher-Jenks algorithm, Jenks Optimisation Method or Fisher exact optimization method. This is a deterministic method to calculate the optimal class boundaries.

Intended compatibility: CPython 3.6+

Wheels are provided via PyPI for Windows / MacOS / Linux users - Also available on conda-forge channel for Anaconda users.

Version Anaconda-Server Badge Build status GH PyPI download month

Usage

Two ways of using jenkspy are available:

  • by using the jenks_breaks function which takes as input a list / tuple / array.array / numpy.ndarray of integers or floats and returns a list of values that correspond to the limits of the classes (starting with the minimum value of the series - the lower bound of the first class - and ending with its maximum value - the upper bound of the last class).
>>> import jenkspy
>>> import json

>>> with open('tests/test.json', 'r') as f:
...     # Read some data from a JSON file
...     data = json.loads(f.read())
...
>>> jenkspy.jenks_breaks(data, n_classes=5) # Asking for 5 classes
[0.0028109620325267315, 2.0935479691252112, 4.205495140049607, 6.178148351609707, 8.09175917180255, 9.997982932254672]
# ^                      ^                    ^                 ^                  ^                 ^
# Lower bound            Upper bound          Upper bound       Upper bound        Upper bound       Upper bound
# 1st class              1st class            2nd class         3rd class          4th class         5th class
# (Minimum value)                                                                                    (Maximum value)
  • by using the JenksNaturalBreaks class that is inspired by scikit-learn classes.

The .fit and .group behavior is slightly different from jenks_breaks, by accepting value outside the range of the minimum and maximum value of breaks_, retaining the input size. It means that fit and group will use only the inner_breaks_. All value below the min bound will be included in the first group and all value higher than the max bound will be included in the last group.

>>> from jenkspy import JenksNaturalBreaks

>>> x = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]

>>> jnb = JenksNaturalBreaks(4) # Asking for 4 clusters

>>> jnb.fit(x) # Create the clusters according to values in 'x'
>>> print(jnb.labels_) # Labels for fitted data
... print(jnb.groups_) # Content of each group
... print(jnb.breaks_) # Break values (including min and max)
... print(jnb.inner_breaks_) # Inner breaks (ie breaks_[1:-1])
[0 0 0 1 1 1 2 2 2 3 3 3]
[array([0, 1, 2]), array([3, 4, 5]), array([6, 7, 8]), array([ 9, 10, 11])]
[0.0, 2.0, 5.0, 8.0, 11.0]
[2.0, 5.0, 8.0]

>>> print(jnb.predict(15)) # Predict the group of a value
3

>>> print(jnb.predict([2.5, 3.5, 6.5])) # Predict the group of several values
[1 1 2]

>>> print(jnb.group([2.5, 3.5, 6.5])) # Group the elements into there groups
[array([], dtype=float64), array([2.5, 3.5]), array([6.5]), array([], dtype=float64)]

Installation

  • From pypi
pip install jenkspy
  • From source
git clone http://github.com/mthh/jenkspy
cd jenkspy/
python setup.py install
  • For anaconda users
conda install -c conda-forge jenkspy

Requirements :

  • Numpy
  • Only for building from source: C compiler, Python C headers and optionally Cython.

Motivation :

  • Making a painless installing C extension so it could be used more easily as a dependency in an other package (and so learning how to build wheels using appveyor / travis at first - now it uses GitHub Actions).
  • Getting the break values! (and fast!). No fancy functionality provided, but contributions/forks/etc are welcome.
  • Other python implementations are currently existing but not as fast or not available on PyPi.