bayesHC1 0.0.1

A package for Bayesian Hierarchical Clustering

Installation Instructions

Run following code in terminal: python3 -m pip install bayesHC1

Then import package and load cython as follows: import bayesHC1

Documentation

bayesHC.bayes_hier_clust(data_vec, alpha, alpha0, beta, kappa, mu)

Given a tuple of datapoints, build a hierarchical cluster tree. Algorithm will continue merging clusters until there is only one overarching cluster. This algorithm is recommended for clustering $<$23 observations.

Parameters: alpha : the expected number of clusters

alpha0 : scale hyperparameter for gamma prior 

beta : location hyperparameter for gamma prior 

kappa : precision hyperparameter for normal prior

mu : mean hyperparameer for normal prior

Returns:nested tuple

Returns a tuple with the following structure: cluster = (c_num,left,right,value,n_k,p_k,d_k)

Where:
    c_num : the number of the cluster
    left : the entire cluster tuple for one subcluster
    right : the entire cluster tuple for the other subcluster
    value : all data values included in the cluster
    n_k : number of datapoints in the cluster
    p_k : the prior on merging
    d_k : a weight on the volume of data in pairs of clusters on the subtree T_k

Example

<<<import pandas as pd <<<df_sim = pd.read_table('simulated_data.csv',delimiter = ",") <<<value = tuple([points] for points in df_sim.loc[:4,'values']) <<<mu, kappa = 0, 100 # mean and standard deviation <<<alpha, beta = 2,10 # scale and location <<<z = bayes_hier_clust(value, alpha, beta, kappa,mu) <<<z

[(9, (4, 0, 0, [24.649825190000001], 1, 1, 2.0), (8, (5, 0, 0, [20.543137890000001], 1, 1, 2.0), (7, (1, 0, 0, [19.404575810000001], 1, 1, 2.0), (6, (2, 0, 0, [4.4017292919999997], 1, 1, 2.0), (3, 0, 0, [10.766117830000001], 1, 1, 2.0), [4.4017292919999997, 10.766117830000001], 2, 0.3333333432674408, 6.0), [19.404575810000001, 4.4017292919999997, 10.766117830000001], 3, 0.25, 16.0), [20.543137890000001, 19.404575810000001, 4.4017292919999997, 10.766117830000001], 4, 0.27272728085517883, 44.0), [24.649825190000001, 20.543137890000001, 19.404575810000001, 4.4017292919999997, 10.766117830000001], 5, 0.3529411852359772, 136.0)]

z[1] (4, 0, 0, [24.649825190000001], 1, 1, 2.0)

Last updated on April 30, 2019.