Hi guys , i am getting a huge difference between cv and lb
CV : 0.21
LB : 0.018
anyone has an idea about this problem ?
When you provide people with your cv/ lb scores, don't forget to mention your validation strategy.
In that way you did, we can't understand and interpret anything. You can also create custom validation strategy : not only random split and kfold...
can you guys share with us a starting kit?
Hi, @bakarysidibe I will give you a hint. Try making the test data similar to sample submission. Then follow the same steps for making train.
PS: it took me 24 hours to prepare train and test itself. So releasing the code would be unethical as this is only a 60 hour hack and some people have spent time to figure out the problem formulation itself.