Runtimes on highly sparse, random LPs with varying numbers of constraints.
The simplex and affine-scaling algorithms were timed against the conic sampling algorithm on random LP problems with n = 100, with 95% sparsity, and with number of constraints k = 128,…,16384. For each k, three problems were generated and timed with all algorithms. Error bars indicate minimum and maximum runtimes. Fully sparse vector data structures were used.