Loading
Cookies help us deliver our services. By using our services, you agree to our use of cookies. Learn more

OpenSNP Height Prediction

OpenSNP


Completed
1275
Submissions
128
Participants
12821
Views

We have a winner !!

Posted by spMohanty about 1 year ago

The challenge has ended, and all the submissions have been regraded against the complete test set !

Congrats to David Baranger for being the winner of this challenge, and for winning a trip to Switzerland to present his work at Applied Machine Learning Days at EPFL.

We Thank each and everyone of you for taking part in this challenge. And we also invite all of you to document your solutions and send them across to be considered for CrowdAI KnowledgeBase.

We look forward to your enthusiastic participation in many more interesting and engaging challenges that we are busy preparing just for you.

Cheers, Mohanty (on behalf of the CrowdAI Team)

2

Posted by SergeKrier  about 1 year ago |  Quote

Hi Mohanty, would you please share the heights of the test set so we can keep improving our model (for fun) ? thanks

1

Posted by spMohanty  about 1 year ago |  Quote

Hi @TeraFlops,

The grader is accepting submissions again. But your submissions will now be marked as “Post Challenge”.

If you have any trouble making submissions, feel free to shoot us an email.

Cheers, Mohanty

Posted by MatSchaeff  about 1 year ago |  Quote

Hello Mohanty,

Thanks for bringing this exciting challenge!

As @TeraFlops, I would like to improve my model.

As written in the Challenge overview the scores have been updated today, they are now computed on 100% of the dataset. Since this update, my old best score dropped a lot. (0.53 ->0.42)

But I just made some post-challenge submissions with the same heights, and I’ve got the same scores as during the challenge (0.53) I think the post-challenge submission still compute automatically the score on the 20%. Is it possible to make it to 100% ?

Cheers, Mat

Posted by spMohanty  about 1 year ago |  Quote

Hi @MatSchaeff ,

Thats true. I did forget to disable the partial grading. That has been fixed now, and the recent submissions regraded. Thanks for pointing it out.

And the drop in R^2 is to be expected when the test set increases, especially in the case when the predictions were overfitting to the exposed part of the test set.

Cheers, Mohanty