Skip to main content

dpcluster is a package for grouping together (clustering) vectors. It automatically chooses the number of clusters that fits the data best based on the underlying Dirichlet Process mixture model.

Project description

Description

dpcluster is a package for grouping together (clustering) vectors. It automatically chooses the number of clusters that fits the data best. Specifically, it models the data as a Dirichlet Process mixture in the exponential family. For a tutorial see “Dirichlet Process” by Y.W. Teh (2010). Currently the only distribution implemented is the multivariate Gaussian with a Normal-Inverse-Wishart conjugate prior but extensions to other distributions are possible.

Two inference algorithms are implemented:

To install locally run:

python setup.py install --user

Usage

Here is a simple example to demonstrate clustering a number of random points in the plane:

>>> from dpcluster import *
>>> n = 10
>>> data = np.random.normal(size=2*n).reshape(-1,2)
>>> vdp = VDP(GaussianNIW(2))
>>> vdp.batch_learn(vdp.distr.sufficient_stats(data))
>>> plt.scatter(data[:,0],data[:,1])
>>> vdp.plot_clusters(slc=np.array([0,1]))
>>> plt.show()

Running this might produce 2-3 clusters depending on the randomly generated data. The adaptive nature of the Dirichlet Process mixture model becomes apparent when we increase the number of data points from n = 10 to n = 500. In this case the clustering algorithm will likely explain the data using only one cluster.

ToDo

  • Implement more clustering algorithms e.g. based on Gibbs sampling, expectation propagation, stochastic gradient descent.

  • Implement more clustering distributions.

  • Re-implement algorithms to take advantage of multi-core or GPU computing.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

dpcluster-0.104.tar.gz (14.6 kB view details)

Uploaded Source

Built Distribution

dpcluster-0.104.linux-x86_64.tar.gz (34.7 kB view details)

Uploaded Source

File details

Details for the file dpcluster-0.104.tar.gz.

File metadata

  • Download URL: dpcluster-0.104.tar.gz
  • Upload date:
  • Size: 14.6 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No

File hashes

Hashes for dpcluster-0.104.tar.gz
Algorithm Hash digest
SHA256 b00d3e7b804e0ab6b80d59d3e9c895e0a3339b0144c3e9db955e65b24719cec4
MD5 29e2f806300fc7c6d150cb0258a5b487
BLAKE2b-256 d6c1ef4c1ee0819cf4721adb2bacdfeeca7492e659f3157dfac47721d62a8cc9

See more details on using hashes here.

File details

Details for the file dpcluster-0.104.linux-x86_64.tar.gz.

File metadata

File hashes

Hashes for dpcluster-0.104.linux-x86_64.tar.gz
Algorithm Hash digest
SHA256 6da1f8579697e35eea48330d6ad6bf2dfae76c016d02d99ae66d3decccb5720c
MD5 7f36cf0c54e2cbb72df1977628b42d7a
BLAKE2b-256 e7e89b669738b8d722a98ba090b2dee2c74a0622e92068af835069ef21cb7588

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page