Optimizes the SVM problem contained in the input model file, saving the result to the output model file. Optimization will continue until either the normalized duality gap (2(p-d)/(p-d), where "p" is the value of the primal objective, and "d" the dual) is smaller than epsilon, or the maximum number of iterations has been exceeded. This implementation handles sparsity using a greedy clustering approach. The small_clusters parameter indicates the size of the clusters: 16 (small) or 256 (not small). Generally, size-256 clusters will give significantly better performance. The active_clusters parameter is the number of clusters which will be active at every point in the greedy clustering algorithm. We have found that 64 works well, but increasing this number will improve the quality of the clustering (at the cost of more time being required to find it). Allowed options: -h [ --help ] display this help -i [ --input ] arg input model file -o [ --output ] arg output model file (may be same as input) -e [ --epsilon ] arg termination threshold -n [ --iterations ] arg maximum number of iterations -s [ --small_clusters ] arg (=0) use size-16 instead of size-256 clusters? -a [ --active_clusters ] arg (=64) number of "active" clusters