Hi all, I use 138 features, and run 5 folds KFold, CV = 0.1705, LB = 0.923. how about you
Hi, I also use 5-fold CV and get 0.8950 locally, 0.8265 on the leaderbord. It seems that your model is overfitting way too much on the training data.
yes I think so, I did adversarial validation but did not find any leakage, will check again, can you please specify objective function or eval_metric you are using if is it ok for you
I tried to directly optimize log loss