Most efficient way to compute

[kenneth-lee-ch]

Given a Bayesnet object, what is the most efficient way in pyagrum to compute the conditional probability P(Y|X) other than using .cpt()?

Currently, I am only doing this

def get_conditional_prob(joint_prob, v_name:set, Condition_v_names:set):
    copy_joint = copy.copy(joint_prob)
    # No condition, return marg prob
    if len(Condition_v_names)==0:
        return copy_joint.margSumOut(list(set(joint_prob.names)-v_name))
    else:
        marg_nominator_names = set(joint_prob.names)-v_name-Condition_v_names
        marg_denominator_names = set(joint_prob.names) - Condition_v_names
        P_denominator = copy_joint.margSumOut(list(marg_denominator_names))
        # No need to marg out
        if marg_nominator_names == set():
            P_nominator = joint_prob
        # Marg out rest
        else:
            P_nominator = copy_joint.margSumOut(list(marg_nominator_names))
        return P_nominator/P_denominator

[kenneth-lee-ch]

I have found the function gum.LazyPropagation(bn) with evidenceImpact()

[phwuil]

evidenceImpact takes care of d-separated variables.

For instance, with the BN A->B->C<-D , evidenceImpact({"C"},{"A","B"})will return P(C|B)since A and C are d-separated given B.