Figure 9: Performance profiles of active learning algorithms. | Scientific Reports

Figure 9: Performance profiles of active learning algorithms.

From: An Evaluation of Active Learning Causal Discovery Methods for Reverse-Engineering Local Causal Pathways of Gene Regulation

Figure 9

The quality of adjacency discovery, accuracy of edge orientation and experimental cost of individual algorithms are used to construct performance profiles of the algorithms. The quality of adjacency discovery was measured by the distance (combined sensitivity/specificity metric) of adjacency discovery. The accuracy of edge orientation was measured by the proportion of correctly oriented edges. Experimental cost was measured by the number of experiments over total number of variables in the dataset. Notice that for edge orientation accuracy, the axis value grow larger when moving away from the origin, for adjacency quality and experimental cost, the axes grow smaller when moving away from the origin. The plot is arranged this way, so that the size of the triangle corresponds to the performance of a particular algorithm. In other words, the larger the triangle, the better performance an algorithm can achieve in terms of the three performance dimensions.

Back to article page