Hi,
I would like to know when the final results for the Optimizing Demand-side challenge are going to be published.
Thanks
ironbar
Hi,
I would like to know when the final results for the Optimizing Demand-side challenge are going to be published.
Thanks
ironbar
Cannot understand what takes so long. I doubt that there are more than 100 participants (LB is not representative), 10 minutes per run (since test set is much smaller) which gives us around 16 hours of computations.
Thanks for your patience. We’re working to execute, review and validate the submissions. We will be announcing the winners as soon as that process is finished, which should be in the near future.
Hello,
We think that our team had a very good approach, but unfortunately we were not careful with the running time, as we realized only 30 minutes after the deadline that we maybe won’t respect the 15 minutes constraint.
We suspect a bug in the way keras clears the model, since model loading was dominating the running time with a battery controller recreated for each simulation. With just the addition of a single line “keras.backend.clear_session()” at the last function call for each simulation, the running time was divided by 4 and became ~5minutes.
So in all cases, would it be possible to know the raw results that each participant achieved on the separate test set, independently of the time limit ?
Hi antoine,
I don’t see you on the leaderboard. Can I ask you which score did you get on the public test set?
Thanks
ironbar
Hi ironbar!
Can you give me your best score on submit test set?
Sure, its my last submission. You can see it with the mouse if you place it on the plot of the scores: -0.1922
Since you ask @ironbar, we reached -0.1985 on the submit dataset, which looked nice especially given that our models had not fully converged yet. Indeed we ran out of time and had to cut the training short at the end to submit our final code.
Hello @antoine @ironbar @duymanhit @ppleskov, thanks again for your patience while we determined the final scores.
We have just announced them here:
https://www.drivendata.org/competitions/53/optimize-photovoltaic-battery/announcements/
Thanks again for participating, we really appreciate your work and feedback!
Hi Bull,
Thanks for the update. Would you also be able to release the set that you used to create the results?
This would be great for research purposes so that we would be able to compare future algorithms with the official results.
I’m wondering if we can publish something related to this competition as a scientific paper.
@bull: Could we use the subject, data, and official results, etc on drivendata.org for research purpose?
Thank you,
Currently the final results are only available for registred users, I think the public leaderboard should be updated because it’s very confusing.
Final scores are publicly available on the “Final Scores” tab here.