Can we use GPU for online evaluation?

NeurIPS 2022: CityLearn Challenge
#1

If yes, how should we specify the environment requirements?

Thanks!

#2

Do you have an answer to that question ? I’m asking myself the same.

#3

The answer seems to be no. We need to use cpu for online evaluation.

#4

Thanks for the question @zheng_shun. I do not know if you can use GPU. Please, @dipam any takes on this.

#5

Hi @zheng_shun

No, we’re not providing GPU access for the Citylearn Challenge. Are you facing issues running any particular model or due to low throughput?

We have tried to set the time limits on evaluation such that its enough, but are open to increasing it in case its needed.

1 Like
#6

@dipam IMHO I think keeping time limits constant throughout the competition is preferable. Changing the time limit hurts anyone who optimized their solution for the current constraints.

  • If the time limit is increased, then some contestants may need to spend a lot of effort redesigning their solution to make the most of the increased time limit.
  • If the time limit is decreased, then some contestants may need to spend a lot of effort redesigning their solution to meet the decreased time limit.

Changes would be less disruptive during the beginning of the competition, but (depending on how you look at it) we’re more than halfway through now.