Figure 5From: Problems with the nested granularity of feature domains in bioinformatics: the eXtasy casePR curves from eXtasy-based experiment. Precision-recall (PR) curves obtained by application of the eXtasy on the data with class balance as in original eXtasy benchmark (left panel) and on the data with realistic class distribution (right panel). Each panel display two solid curves - the one corresponding to the standard Random forest classifier training with bootstrapping and the one corresponding to hierarchical sampling based training. The shaded areas represent empirical 95% confidence intervals obtained as a by-product of threshold averaging.Back to article page