2019/05/29 by Anthony C. Constantinou, Constantinou, Anthony C.
Computer Science · Decision Sciences · #Bayesian Modeling and Causal Inference #Advanced Graph Neural Networks #Multi-Criteria Decision Making
paper · pdf · doi:10.48550/arxiv.1905.12666
Several structure learning algorithms have been proposed towards discovering\ncausal or Bayesian Network (BN) graphs. The validity of these algorithms tends\nto be evaluated by assessing the relationship between the learnt and the ground\ntruth graph. However, there is no agreed scoring metric to determine this\nrelationship. Moreover, this paper shows that some of the commonly used metrics\ntend to be biased in favour of graphs that minimise edges. While graphs that\nare less complex are desirable, some of the metrics favour underfitted graphs,\nthereby encouraging limited propagation of evidence. This paper proposes the\nBalanced Scoring Function (BSF) that eliminates this bias by adjusting the\nreward function based on the difficulty of discovering an edge, or no edge,\nproportional to their occurrence rate in the ground truth graph. The BSF score\ncan be used in conjunction with other traditional metrics to provide an\nalternative and unbiased assessment about the capability of a structure\nlearning algorithm in discovering causal or BN graphs.\n