On Mon, May 19, 2014 at 11:08 AM, Dmitriy Lyubimov (JIRA) <[email protected]>wrote:
> [~avati] do you think you could perhaps explain (or reference principled > foundation publication) of the algorithm that is happening here? One of the most commonly effective compression techniques is dictionary + run-length. For instance, the binary matrices that much of our software uses would have massive compression using this. For instance, a binary vector with 1million elements with 0.01% sparsity would compress to about less than 200 bytes using these techniques and a very naive implementation. Our current sparse representation requires about 1200 bytes.
