{"id":1010,"date":"2021-06-03T06:07:29","date_gmt":"2021-06-03T06:07:29","guid":{"rendered":"http:\/\/jsr.isrt.ac.bd\/?post_type=article&p=1010"},"modified":"2021-06-03T06:08:47","modified_gmt":"2021-06-03T06:08:47","slug":"searching-across-markov-equivalent-directed-acyclic-graph-models","status":"publish","type":"article","link":"http:\/\/jsr.isrt.ac.bd\/article\/searching-across-markov-equivalent-directed-acyclic-graph-models\/","title":{"rendered":"Searching Across Markov Equivalent Directed Acyclic Graph Models"},"content":{"rendered":"
Learning the structure of a process that can be represented by a directed acyclic graph (DAG) based on data alone can be a challenging problem because many graphs may encode the same conditional independence relations. However, searching across equivalence classes can greatly reduce the search space, thereby making the search more e\ufb03cient. This paper presents the DECS algorithm, which is an extension of Edwards and Havernack\u2019s EH-procedure (Edwards, 1995) for undirected graphs to DAG equivalence classes. We also provide necessary graphical criterion for the DAG submodel relation and prove its su\ufb03ciency in special cases. This criterion facilitates the moves made across equivalence classes in the search space. Finally, the DECS algorithm is demonstrated on real data sets.<\/p>\n