Thanks for the host and the great participants in this challenge! The dataset is high quality, distribution looks reasonable and seem to closely match real display ads data. We don't have any major issue when experimenting various ideas. Participating in this challenge provided great learning opportunity for me to actually play with various popular ML toolkits. We've been trying various kind of solutions and the best performing one on the leaderboard is actually surprisingly simple in general. Other fancy techniques we tried to stack on top of it tend not to provide as much lift as the vanilla. Looking forward sharing more details in the workshop.
Team i486DX2
Posted by: guruqu @ Aug. 4, 2021, 7:57 a.m.