Is it to train with five data sets separately, then build different models, and then test? Are these five model relationships parallel?Posted by: lhg1992 @ Aug. 10, 2018, 8:59 a.m.
Reading this (https://competitions.codalab.org/competitions/19836#learn_the_details-evaluation) should be helpful to your question.Posted by: xiayunsun @ Aug. 10, 2018, 9:59 a.m.
I have browsed the webpage many times according to your suggestion. I think the example diagram in the webpage only shows a dataset, such as AA, so can you assume that the five datasets are independent of each other? But what about the same training framework for these datasets?Posted by: lhg1992 @ Aug. 10, 2018, 11:52 a.m.
We are using 5 data sets, each of which evaluated in the same scenario. This is an AutoML competition, so we look for a method that performs well for many data sets. The goal of having 5 data sets is for the results to be more conclusive.