effici causal discoveri algorithm linear model zhenx wang laiwan chan bayesian network learn algorithm wide causal discoveri pioneer work exist algorithm phase depend analysi algorithm tpda effici sens polynomi time complex limit improv tpda depend mutual inform base condit independ ci test easi appli continu data addit tpda phase approxim skeleton bayesian network effici practic paper propos phase algorithm partial correl base ci test phase algorithm construct markov random field data close approxim structur true bayesian network phase algorithm remov redund edg ci test true bayesian network show phase algorithm partial correl base ci test deal continu data arbitrari distribut gaussian distribut 