next up previous contents
Next: Imbalanced Class Distributions: Up: Other Properties and Open Previous: Parallel Implementations:   Contents

Memory Requirements:

It is also clear that each individual binary task in a round-robin binarization has fewer training examples than the original tasks. For multi-class tasks that are too large to be performed in memory, pairwise classification may provide a simple means to reduce the size of the learning task without resorting to subsampling. Note that this is not the case for unordered class binarization or error-correcting output codes.



Johannes Fürnkranz 2002-03-11