I finished testing xz, zpaq, and nncp on different extraction parameters. All of them (unsurprisingly) demonstrated no compression. Periklis mentioned that my results may have differed from the original paper, since the input size I used were much bigger. I’m not really sure how I could use smaller input sizes though, since I picked the bare minimum data size such that the NIST tests are still accurate (some tests require quite a bit of data).
On a tangent, actually getting nncp to run on Amarel (or any computer for that matter) is such a pain. It requires a GPU to complete any task at a reasonable speed, and LibNC (dependency) seems to fall apart on the latest release of CUDA on my Arch machine. Additionally, only a few of the GPU clusters on Amarel work with this library, so I had to play a guessing game to hopefully get sessions with the required hardware.
Aside from that, spent most of the week thinking about how to present my menagerie of results, theoretical and practical, in the final report.