BN.py

def first_number_index(L):
    """
    function to find the index of first number in one line of document
    """
    for i,s in enumerate(L):
        try:
            float(s)
            return i
        except:
            pass
        
def file_to_cpt(filename):
    """
    function to convert text file into conditional porbability table and compute compactness
    """
    with open(filename,'r') as f:
        # conditional probability table
        CPT = {}
        # nodes in BN
        nodes = []
        # num of CPT lines
        n_rows = 0
        for line in f:
            line = line.replace('\r\n','')
            if line != 'END':
                content = line.split()
                print 'Reading node '+ content[0]
                nodes.append(content[0])
                # index of first number
                index = first_number_index(content)
                parents = content[1:index]
                if 'NONE' in parents:
                    n_rows += 1
                else:
                    n_rows += 2**len(parents)
                CPT[content[0]] = {'parents':parents,'cpt':[float(num) for num in content[index:]]}
                #cpt = [content[0],parents,[float(num) for num in content[index:]]]
                #CPT.append(cpt)
            else:
                print filename+' read.'
        compactness = 1.*n_rows/(2**len(nodes))
        return CPT,n_rows,2**len(nodes),compactness,nodes
            

def bool_to_var(Bool,Var):
    """
    function to transform boolean list into list with variable name
    e.g., Bool = [1,0], Var = ['Burglary','Earthquake'], then the output would be ['nBurglary','Earthquake']
    """
    new_var = []
    for i,j in enumerate(Bool):
        if j == 1:
            new_var.append('n'+Var[i])
        else:
            new_var.append(Var[i])
    return new_var

def parents_to_index(parents,jd):
    """
    function to use parents to get index in cpt
    """
    binary = ''
    for parent in parents:
        if parent in jd:
            binary += '0'
        else:
            binary += '1'
    return int(binary,2)
    
    
def joint_prob(CPT,nodes,out_filename):
    """
    function to compute full joint probability and write result to a txt file
    """
    # num of multiplications
    n_mul = 0
    # num of additions
    n_add = 0
    # list to store joint distributions
    result = []
    for i in range(1<<len(nodes)):
        # joint probability
        joint_prob = 1.
        # True/False boolean value
        tf = bin(i)[2:]
        tf = '0'*(len(nodes)-len(tf))+tf
        tf = map(int,list(tf))
        jd = bool_to_var(tf,nodes)
        for node in jd:
            if node[0] != 'n':
                parents = CPT[node]['parents']
                if 'NONE' in parents:
                    joint_prob *= CPT[node]['cpt'][0]
                    n_mul += 1
                else:
                    index = parents_to_index(parents,jd)
                    joint_prob *= CPT[node]['cpt'][index]
                    n_mul += 1
            else:
                parents = CPT[node[1:]]['parents']
                if 'NONE' in parents:
                    joint_prob *= (1.-CPT[node[1:]]['cpt'][0])
                    n_mul += 1
                    n_add += 1
                else:
                    index = parents_to_index(parents,jd)
                    joint_prob *= (1.-CPT[node[1:]]['cpt'][index])
                    n_mul += 1
                    n_add += 1
        result.append([jd,joint_prob])
        with open(out_filename,'w') as f:
            for line in result:
                f.write(str(line[0])+'\t')
                f.write(str(line[1])+'\n')
            
    return n_mul,n_add

def p_summary(filename,CPT,n_cpt,n_jd,compactness,n_mul,n_add):
    """
    function to print summary statistics for BN
    """
    print CPT
    print '------------------------------------------------------'
    print 'Bayesian Network                  :\t'+filename
    print 'Num of lines in joint distribution:\t'+str(n_jd)
    print 'Num of CPT lines                  :\t'+str(n_cpt)
    print 'Compactness                       :\t'+str(compactness)
    print 'Num multiply operations           :\t'+str(n_mul)
    print 'Num add operations                :\t'+str(n_add)
    print '------------------------------------------------------'
    print '\n'

def comp_joint_dist(filename):
    # construct CPT from source file
    CPT,n_cpt,n_jd,compactness,nodes = file_to_cpt(filename)
    # file to store full joint distribution
    out_filename = 'out_'+filename
    # compute full joint distribution
    n_mul,n_add = joint_prob(CPT,nodes,out_filename)
    # print summary information
    p_summary(filename,CPT,n_cpt,n_jd,compactness,n_mul,n_add)

filename = ['examplebayes.txt','bayesnets1.txt','bayesnets2.txt','bayesnets3.txt','bayesnets4.txt']
for name in filename:
    comp_joint_dist(name)