memebayes 0.2.1

A bayesian network implementation

Bayesian Network

Public library for a Bayesian Network in python. This library uses only re as dependency. It works as a standalone library and can be used in any python project.

It is defined as a class BayesNetwork which has the following methods:

• add_node : Adds a node to the network
• edit_node : Edits a node in the network
• compact : Shows a compact representation of the network
• defined : Checks if every node is defined
• query : Queries the network for a given node

Installation

pip install memebayes

Usage

Importing the library

from memebayes import BayesNetwork

Creating a network

network = BayesNetwork(nodes:list)

• nodes is a list of strings, with the names of the nodes in the network.

Adding a node

network.add_node(node:str, parents:list, probabilities:dict)

• node is the name of the node to be added.
• parents is a list of strings, with the names of the parents of the node.
• probabilities is a dictionary with the probabilities of the node. The keys are the possible values of the node, and the values are the probabilities of the node given the parents. If the node has no parents, the keys are the possible values of the node, and the values are the probabilities of the node.

Editing a node

network.edit_node(node:str, parents:list, probabilities:dict)

Compacting the network

network.compact() # Returns a string Ej.P(C)P(B|C)P(A|B,C)

Checking if the network is defined

network.defined() # Returns a boolean

Querying the network

network.query(query:str) # Returns the probabilities of the node given the evidence

• query is a string with the query to be made. The format is node|evidence, where node is the name of the node to be queried, and evidence is a list of strings with the evidence of the query. The format of the evidence is node=value, where node is the name of the node, and value is the value of the node. If there is no evidence, the format is node.

Examples

from memebayes import BayesNetwork

net = BayesNetwork(["B","E","A","J","M"]) # Create a network with the nodes B, E, A, J and M

net.edit_node("B",pt={"B":0.001}) # Edit the node B with the probability of being true
net.edit_node("E",pt={"E":0.002}) # Edit the node E with the probability of being true

net.edit_node("A",parents=["B","E"], pt = {"A|BE":0.95,"A|B-E":0.94,"A|-BE":0.29,"A|-B-E":0.001}) # Edit the node A with all the possible probabilities of being true given the parents B and E

net.edit_node("J", parents = ["A"], pt = {"J|A": 0.9, "J|-A": 0.5}) # Edit the node J with the probability of being true given that A is true
net.edit_node("M", parents=["A"], pt={"M|A": 0.7, "M|-A": 0.01}) # Edit the node M with the probability of being true given that A is true

print(net.compact()) # Print the compact representation of the network (Ej.P(B)P(E)P(A|B,E)P(J|A)P(M|A))
print(net.defined()) # Check if the network is defined (True)
print(net.query("B")) # Query the network for the node B (0.001)
 
print(net.query("B|JM")) # Query the network for the node B given that J and M are true (0.0007)

References

This library is based on the following references:

• https://en.wikipedia.org/wiki/Bayesian_network
• http://courses.csail.mit.edu/6.034s/handouts/spring12/bayesnets-pseudocode.pdf
• https://www.cs.cmu.edu/~awm/15781/slides/bayesinf05a.pdf