Probabilistic Inference of Biological Networks via Data Integration

Figure 4

Mean test error as a fraction (-axis) versus the number of patterns learnt (-axis) for the top-performing pairwise kernel, . Error bars depict a 95% confidence interval for 5-fold cross-validation test error averaged over 10 distinct data subsets, each with . The upper curve gives the performance if we learn all the data sequentially (from a common start set) in random order. The lower curve gives the test accuracy if the next addition to the training set is chosen based on having the lowest confidence predicted link-label.