Fig. 3

Test-set error for different classifiers, number of variables and boosting iterations. The figure reports the average test set error as a function of the number of boosting iterations for different number of variables (p=100, 1000 and 10000). The difference between the classes was moderate, the correlation structure was exchangeable and there were 10 variables per block and 100 differentially expressed variables, see the “Methods” section for more details