I just checked my other submissions. After the competition ended it seems there has been a reevaluation of all the submissions done and the submissions that were scored lower than those with highest score I had achieved during competitions, got higher score. For ex: a submission with score 0.587 during competition scored 0.658. This score puts me in the top 20. But since I had not marked that particular submission for consideration on leaderboard, it won't be considered. Why did the competition evaluate my unmarked submission?
As submissions are made, they are evaluated overall on both public and private LB. Only public LB score is shown during competition. After the competition, private LB score is shown for all submissions. It's automatic.
Then shouldn't the private leaderboard consider all the submissions instead of considering only 2 marked by the participants? I am ranked lower but I have other submissions which will bring me up higher up the leaderboard.
Yes, any submission not ticked is considered only an experiment. You have to put forward your 2 selected models you wish to be judged on. If Zindi has to pick submissions automatically based on Private LB, the score could be by chance and not necessarily a model with confidence by the submitted user. In real life, we make many experiments but choose what model to be deployed ourselves. Any decision on the choice of model has a major impact, so we need to have some confidence in our generalization and our choices. Zindi evaluates us on choices in that regard.
Yes i concur. More or less Zindi is trying to groom and build every of its participant on how to make a right and very valuable choice when it comes to model deployment in real life scenario or setting where there will be no Leaderboard or platform like Zindi to rank or score our model. Henceforth, one's intuition must not be hindered with doubt, our confidence on any kind of final model selectred by us must be intact.