5 ms·
Welcome to benchmarklandia, where you find the top algorithm in an area and write a stripped down version that outcompetes it on a different set of data and get
by fnord123 6y ago
Welcome to benchmarklandia, where you find the top algorithm in an area and write a stripped down version that outcompetes it on a different set of data and get good results.
See: hadoop->spark->spark with infiniband->spark with nvme->timely dataflow on a laptop; csv parsers that don't support scientific notation floats, etc. I'm sure others can give moren interesting examples.
- albertwang 6y agoOut of (self-interested) curiosity, what are you referring to with the "nvme->timely dataflow on a laptop" reference? The closest benchmark I could find was the "FASTER State Management for Timely Dataflow" paper from ETH Zurich, but that wasn't run on a laptop.
- fnord123 6y agohttp://www.frankmcsherry.org/assets/COST.pdf http://www.frankmcsherry.org/assets/COST.pdf
- frankmcsherry 6y agoIt's a bit different, right, because that paper presents better algorithms on the same datasets as the work it cites.
- fnord123 6y agoYes it's different. Another part of Benchmarklandia is where researchers make custom algorithms for the same dataset. Benchmarks targeting MNIST is a great example. :)