Are people still having disparity between LB and CV? My CV has gone up to 0.91 but lb score still plateau-ing at 0.68. Seen many people break the 0.7 mark so curious how people got over this disparity
1. Some of the people who have lb scores over 0.7 just have models that are highly overfit and their private scores will be lower
2. They genuinely have better models and found a clever way to account for the differences between the distributions of Train.csv and Test.csv as well as the imbalance in the data
Experiencing the same here.
2 theories:
1. Some of the people who have lb scores over 0.7 just have models that are highly overfit and their private scores will be lower
2. They genuinely have better models and found a clever way to account for the differences between the distributions of Train.csv and Test.csv as well as the imbalance in the data
Quite the conundrum, whether to trust the model with better Cv scores or lb scores
you h've got two submissions 😉
Haha true, will just submit the best of both and hope for the best
The best advice is to trust your approach because the shake up is going to be crazy once the comp is over
it is just Good cross validation method solves a lil bit of this problem. for me my CV is 0.82 and LB is 0.72, it correlates well i guess