What feature would you like to see Kaggle work on?

[paper mica]

Give us some feedback, tell us what features would make Kaggle work better for you!

[near sinew]

Dark theme on main kaggle page 🙂

[shadow tide]

I know it’s too much, but it would be awesome to have 1 scheduled notebook with GPU, just in medal competitions.

[wintry basin]

@shadow tide can you elaborate a bit more on why you'd find this useful?

[shadow tide]

Hi Will, sure. I think it could be useful since in many competitions the training times are quite long at times, and sometimes the notebook execution fails due to being open for so many hours, it happened to me several times in the Contrails competition, 5 to 10 hours of train and in the end, It didn't save the models checkpoint due to the "failed to save draft" error, so I ended up wasting a lot of GPU hours, mines and Kaggle in general.
In addition, being able to run a notebook in scheduling would allow us to run it without having to have the laptop running a train all day in parallel while you work, with the risk of chrome freezing, because you are doing many things at the same time, or some OS update in the background, internet connection issues, things like that. In the end for that kind of things, you end up wasting many GPU hours.
I also imagine that it would allow us to try more things, since you change the configuration parameters, and then leave it to run later, go back to work, and then check when you receive the email that your notebook was executed. I perfectly understand that it is not something easy, and Kaggle gives us too many things for free, just a suggestion.

[paper mica]

To confirm, are you running it just within the editor or with the "Save Version > Run All"? If I'm understanding right, the latter should do what you need. It'll start off a new session and you can close the editor and leave the session to run in the background.

[shadow tide]

Editor

[paper mica]

okay so you should be able to use the "Save Version" button in the top right and make sure "Save & Run All (Commit)" is selected. It should do what you want

[shadow tide]

Thank you Nathan. 2 years on Kaggle and I didn’t know you could do that, I always thought that if you closed Kaggle website, that train would stopped 🤦. Sorry for the inconvenience.

[paper mica]

No worries! There's lots to the platform. Just glad the problem could be solved for you

[late eagle]

Discussion posts are the oldest things on Kaggle to have voting and from the very beginning have had upvotes and downvotes (originally all private). Eventually we made it possible to see upvotes, but kept downvotes private to respect the anonymity of those who choose to disagree with other people. Downvotes are a useful tool for helping other readers understand when content is unhelpful or unpopular which is why they exist for discussions.

Notebooks, Datasets, Models, etc. are pretty different - they represent artifacts/resources that usually had a lot of time put into them. Upvotes here typically mean "this is good" vs. "I agree with this" and in this context downvotes would be a lot less useful. For inappropriate content we rely on reports rather than downvotes.

Over the years we have seen a lot of debate around votes (wanting to add downvotes to notebooks, remove downvotes from discussion, make downvotes public etc.) A lot of this tends to stem from the progression system and people caring more about votes because of it. We'd rather work on tweaking how the progression system influences what people care about than tweak voting itself, but we are always reading threads and feedback and we'll never say never to considering an idea.

One thing I will say is that I've seen many people over the years convinced that there are people out to get them who are consistently downvoting their stuff and that this is a problem on the site. In every case we've ever looked into this has never actually been the case - it's natural to be paranoid perhaps, but actual downvote vendettas are very rare (and from a progression point of view we actually have systems that stop them from counting, they work in an inverse ways to the systems we have to stop counting having one friend upvote all your stuff).

[paper eagle]

It would be great to be able to delete a notebook from within the notebook, maybe a File->Delete Notebook/Move to Trash like Google Colab.
How's this helpful?
Atleast for me, I often create new notebooks to try things out quick and then dispose of them, while working on a main notebook alongside. It's a little tiresome to go to the 'Your work' page every time and then find the notebook I want to delete by name, and then delete it.
It would be convenient to have an option in-notebook as well, to delete it, once you're done with it.
Thanks!

[paper mica]

I've passed this idea along to the team. A workaround I sometimes use it to just have one "scratchpad" notebook that I go back to and then just delete cells one I'm done with it. You can also have it so the first cell is kept around and does all the imports/initializations you find yourself using

[paper eagle]

Thank you!
And that's a great idea, I use that too!
But sometimes I want to simultaneously test things out with different versions of packages installed.
Or other cases like, HF transformers requires restarting the kernel on installing a related package (datasets) after initialising and using transformers due to cached values during transformers init process.

These are probably niche, so it's not an issue.

Thanks for your help and for passing along the idea!

[fluid salmon]

I would like to see functionality where, if a submission fails to find a CSV file it should not count towards daily submissions. I have followed all of the competition rules and read all of the documentation and still cannot figure out what's going wrong and I have wasted a whole day because I ran out of daily submissions trying to debug the issue.

[vestal notch]

Deployment feature as in huggingface would be very helpful

[paper eagle]

To be able to import notebooks originally created in colab.
A notebook that was created on colab seems to have issues being imported on kaggle due to some metadata differences?

After importing my notebook, on quicksave it currently fails, erroring out at
[NbConvertApp] Converting notebook __notebook__.ipynb to html

I had imported the notebook from my GitHub using the 'Import from GitHub' feature.

[ripe zinc]

Changing Username

[sonic badger]

Ability to make multiple scores in a competition, current neurips unlearning contest would benefit greatly from this imho

[marsh minnow]

It would be nice to be able to see scoring time for each submission, because it is not always the same as running time of the notebook.