Yunseong Lee
Yunseong Lee
The dataset in our Eurosys evaluation was synthetic, generated by a script included in Petuum. The synthetic dataset was helpful to understand basic trend as we can easily change the...
Dolphin has not considered much about fault tolerance. To handle the failed Task/Contexts, we need to 1) clarify that which states must be maintained, and 2) create a checkpoint mechanism...
We can use HTrace for performance profiling, especially for finding the source of bottleneck. This issue aims at applying HTrace on Dolphin, to the points we're interested with.
For now, the App does not use "EM" and the computation time at Trainer is fixed over iteration. We can extend the application, so that the computation and communication time...
In #535, we're planning to shutdown the entire job when an Exception occurs at Evaluator-side. We concluded to take this approach for making tests and development easier, but we should...
In MLRWorker's constructor, there are a lot of parameters for algorithms, but comments are missing. Though the named parameters have their `doc` (written by @jsjason), it'd be great to see...
Overall codebase in Cay actively throws `RuntimeException`. It is useful for debug as the job fail fast if bugs exist, but sometimes we need to figure out which Exception was...
We've run experiments by using same LR algorithm with URL reputation dataset on multiple frameworks: Dolphin, Vortex. As @beomyeol mentioned at the meeting, we've seen some performance issues such as...
This is part of #265. As described above link, LR algorithm does not seem to provide enough accuracy. We need to verify that the accuracy increases to a reasonable degree...
This is part of #265 @gyeongin and @beomyeol reported that Vector computation in Dolphin was too slow. First thing we can try is replacing vector library rather than using Mahout's....