bayesian-cut 0.1.2b0

Installation

• Install the package for Python3 using the following command. Some dependencies might require an upgrade (scikit-learn, numpy and scipy).

  $ pip install bayesian_cut
• Now you can import the bayesian cut class by typing

  from bayesian_cut.cuts import Model

Quick Start

Use the bayesian_cut package for performing graph cuts

# Load the package
from bayesian_cut.data.load import load_data
from bayesian_cut.cuts.bayesian_models import Model

# Load an example network
network_name = 'karate'
X, Y = load_data(network=network_name, labels=True, remove_disconnected=True)

# Set the parameters for the model
n_samples_per_chain = 75
n_chains = 15
C = 2
model_params = {
    'alpha_in': 1e-2,
    'beta_in': 1e-2,
    'alpha_out': 1e-2,
    'beta_out': 1e-2,
    'b': 1,
    'gamma': 3
}

# Define the model
BC = Model('ShiftedApproximateBayesianCommunityDetection',
            X,
            model_params,
            Y=Y,
            C=C,
            block_sampling=False,
            marginalize_phi=True
            )

# Add the number of chains to run
BC.add_chains(number_of_chains=n_chains)

# Run the Gibbs sampler
BC.run_chains(n_samples=n_samples_per_chain,
              n_prior_updates=20,
              verbose=2,
              save_all_samples=False,
              parallel=True
              )

# Obtain the cut with the highest log-likelihood
graph_cut_z_vector = BC.get_best_chain().max_log_lik_z_

# Plot the cut as an adjacency matrix
from bayesian_cut.utils import utils
utils.cluster_plot(BC)

# Done