What is the latest date of merging teams?
Before the beginning of round 3 or before the beginning of final submission round?
Just curious, because clearly a late team-up can double the compute hours.
Created by Yuanfang Guan ???? yuanfang.guan Dear Yuanfang,
I agree with the fact that compute hours can be a limiting factor. For example for the caffe pipeline just preparing the data takes ~2-3 days without any data augmentation or other preprocessing methods. If we further manipulate the data, just preparing the training dataset can take ~5 days, which I would say is a bit too much compared to the allocated compute hours. Of course you can do this only once, but if you want to investigate multiple preprocessing approaches than you are limited, as your remaining time is not enough to also test multiple architectures.
Best,
Anamaria Dear Yuanfang,
Thank you for bringing this up. We are discussing this internally and hope to have an answer for you soon.
Best,
Thomas OK... just to clarify that in general i am fine with anything others are doing, as long as my resource is not affected.
i just felt that i (and perhaps several others) am the only one being limited by compute hours here. why NOBODY else complained about compute hours? i agree that many people don't need that many hours. but **NOBODY** else complained, don't you think it is weird? This is because one can use a pseudo account. Or one can ask a lab member to register an account and continue to test, **which doesn't break any rule. Or split teams. **But i have been real named for such a long time that i can't.
rules are applied only to those who obey them.
so you can just triple the compute hours. then, you can allow teams to split and merge infinite time as one likes. Dear Yuanfang,
I have forwarded your concern to Gustavo and Justin. Thanks in advance for your patience.
Best,
Thomas yes, please forward to thomas schaffter and justin guinny
as you know, i don't work in teams.
then, other teams have a huge advantage, because they can split even in the 3rd round and double and even triple their hours, and then combine in the final round. A team of 5 people can practically have 1800 hours, while I only have 350! They have 15 submission tests, while I have 3.
I suggest three ways to work this out:
1. **no team-up should be allowed after the start of the 3rd round. What could possibly be the reason that until now one doesn't know who he wants to team up, but wait until the final submission? other than they want to exploit hours and test submissions?**
2. **Or you give me 1000 hours. Even if you give me 1000 hours, it is barely enough for me to use. As long as you give me enough hours, I don't care what other people are doing or that they can make 50 test submissions**
3. **Or for any team that formed not at the beginning of the challenge, a thorough check of docker file must be made to ensure that prior to their team-forming date, no private code-base is shared. Otherwise, one cannot explain why people from originally different teams using the same code.** Dear Yuanfang,
I think you can merge teams every round as long as you only make submissions with that team per round(I will confirm with other challenge organizers).
If you submit as part of a new team then all members of that team will only get to submit as that team which would then keep the compute hours the same. If anyone on that team submits as an individual, none of the other members on the team will be able to submit as that team.
If 2 teams decide to work together but keep their original teams, and one of the 2 teams becomes the best performer, only members of the best performing team will be recognized regardless of the cooperation. I hope this makes sense.
Best,
Thomas