I was wondering whether it is possible to have more daily submission available.
I generated a training and validation set from the dataset provided. However I currently have a significant discrepancy between the accuracy I achieve on my validation (higher than 90%) and the one I get with the test set (around 70%).
Unfortunately with only 5 submissions a day it is hard to make many attempts in order to fix it.
15 or 20 submissions a day would really help.
@AmaurySu: I acknowledge the argument, but the low per-day submission count is by design.
If we provided a much larger number of submissions, then in the process of trying to fix the divergence in the validation set scores and training set scores, you will end up overfitting to the test set.
In this case, you should try ensuring that multiple training-validation splits at your end have consistent scores, and then ideally a similar score should translate as the score on the test set.