dp Package

madbayes Package

Implements some of the MAD-Bayes algorithms by Tamara et al.

dpmeans Module

The DP-means algorithm by Kulis et al.

infpy.madbayes.dpmeans.dpmeans(x, lambda_, progress_plots=False)

The DP-means algorithm by Kulis et al.

Parameters :

• \(x\) : input data, a sequence of length \(N\)
• \(\lambda\) : cluster penalty parameter
• progress_plots : Scatter plot clusters at every iteration

Returns:Cluster indicator variables

Algorithm:

1. Initialise:

• Number of clusters \(K=1\)
• Global cluster mean \(\mu_1 = \frac{1}{n} \sum_n x_n\)
• Cluster indicator variables \(z_n = 0 \quad \forall n\)

1. Repeat until convergence:

• For each point \(n\):
  • Compute distance to each cluster \(d_{nk} = ||x_n - \mu_k||^2\)
  • If \(\min d_{nk} > \lambda\) then set \(K=K+1, z_n=K, \mu_k=x_n\)
  • Otherwise set \(z_n= \arg\!\min_k d_{nk}\)
• For each cluster \(k\), compute \(\mu_k = \frac{1}{|\{n: z_n = k\}|}\sum_{n: z_n = k} x_n\)

infpy.madbayes.dpmeans.plot_clusters(x, z, format_cycler=None)