ABSTRACT
This chapter shows that causal relations are represented by a directed graph. It provides an overview of the major kinds of input data, background knowledge, and algorithms, along with descriptions of the problems that arise and strategies for dealing with those problems. In contrast to constraint-based searches, score-based searches look for a causal model that maximizes or minimizes the score of a causal graph relative to a data set. The chapter considers constraint-based searches under a variety of different background assumptions. Typically, causal search algorithms fall into one of three broad classes—constraint-based, score-based, or hybrid algorithms. Constraint-based algorithms treat causal search as a constraint satisfaction problem. The Parents and Children algorithm runs quickly on sparse graphs, since in those cases it only performs independence tests conditional on sets of small size. In practice, statistical tests are performed by the PC algorithm to determine if a conditional independence relation is judged to hold.
