Hybrid Bayesian Networks

Network initialization

If the dataset contains both discrete and continuous variables, HybridBN is recommended to use. To initialize a HybridBN object, you can use the following code:

from  bamt.networks.hybrid_bn import HybridBN

bn = HybridBN(has_logit=True, use_mixture=True)

HybridBN has two additional parameters has_logit and use_mixture. has_logit is used to determine whether to use logit nodes. Logit nodes use machine learning algorithms to represent variable. Logit nodes are discrete nodes that have continuous root nodes; classification models are used to model conditional distributions in such nodes.

Data Preprocessing

Before applying any structure or parametric learning, the data should be preprocessed as follows:

import bamt.Preprocessor as pp
import pandas as pd
from sklearn import preprocessing

data = pd.read_csv('data.csv')

encoder = preprocessing.LabelEncoder()
discretizer = preprocessing.KBinsDiscretizer(n_bins=5, encode='ordinal', strategy='quantile')

p = pp.Preprocessor([('encoder', encoder), ('discretizer', discretizer)])
discretized_data, est = p.apply(data)

info = p.info

Structure Learning

For structure learning of Hybrid BNs, bn.add_nodes() and bn.add_edges() methods are used.

from pgmpy.estimators import K2Score

bn.add_nodes(info) # add nodes from info obtained from preprocessing

params = {
            # Defines initial nodes of the network, list of node names
           'init_nodes':[...]
            # Defines initial edges of the network, list of tuples (node1, node2)
           'init_edges':[...]
           # Strictly set edges where algoritm must learn, list of tuples (node1, node2)
           'white_list':[...]
           # blacklist edges, list of tuples (node1, node2)
           'bl_add':[...]
           # Allow algorithm to remove edges defined by user, bool
           'remove_init_edges':True
          }

# Structure learning using K2Score and parameters defined above
bn.add_edges(discretized_data, scoring_function=('K2', K2Score), params=params)

bn.plot('foo.html') # add nodes from info obtained from preprocessing

Parametric Learning

For parametric learning of continuous BNs, bn.fit_parameters() method is used.

bn.fit_parameters(data)

bn.get_info() # get information table about the network