Based on a discussion with @steincaleb, I created a quick-and-dirty figure of classification accuracies that summarizes things quite nicely.
The classification error rates are obtained with 1000 reps. Standard errors are approximately 0.01 (difficult to show on the plot). The variables are selected using the blocking idea we discussed, so the results have a natural interpretation as truly having more information.
It's quite interesting to see that GRDA is hardly effected by the dimension of the data!
Based on a discussion with @steincaleb, I created a quick-and-dirty figure of classification accuracies that summarizes things quite nicely.
The classification error rates are obtained with 1000 reps. Standard errors are approximately 0.01 (difficult to show on the plot). The variables are selected using the blocking idea we discussed, so the results have a natural interpretation as truly having more information.
It's quite interesting to see that GRDA is hardly effected by the dimension of the data!