playground-series-s3e20

[fervent moat]

Hi!

[grave ivy]

@fervent moat Is this your first playground series competition, or have you done multiple? What do you think of them?

[fervent moat]

This is my 7th

Have been doing every single one since S3E14

[fervent moat]

They are great! I remember one of the users posted in the forums regarding the lack of variety of challenges and y’all have responded very well to it

[narrow cliff]

Thanks, that's great feedback. I'll let the competitions team know 🙂

[fervent moat]

Can we trust the postprocessing this time

It turned out really badly in the ICR competition

[zinc perch]

A general answer to that question is no. Make that a NO.

[fervent moat]

What is OOD

oh ok

[hard dune]

Hello I am having problem while submitting my solution, can anyone help or suggest me where it's going wrong, I am new in Data Science and Machine Learning field. Please feel free to DM me.
Also want to ask to admin , can I share my notebook link here?

[grave ivy]

Please share notebook links in #🔗┊sharing-projects, to get help on making a submission you might need to be more specific about what issue/error you are facing.

[hard dune]

Thanks @grave ivy

[carmine kestrel]

I'm observing single digit CV rmse with "poisson" regression that isn't appearing as good on leaderboard scores. Are you all going with your best 2 CV, in spite of leaderboard standing? (I'll put together a post for public discussion, too)

[crude wind]

As the current rank#1 on public LB right now, I would say EDA from visualizations and figure out error sources are far more important than modeling😆

Cuz I've worked on time-series data for years, I am very sensitive to the patterns and behaviors of human-related data

[fervent moat]

But probing the LB can turn out to be disastrous as seen in the ICR competition

[stray dew]

Has activity on kaggle reduced largely due to this??
I could sense far lower posts on kaggle forums now a days

[plain lake]

It’s the middle of august

[zinc perch]

I think the activity on Kaggle has dropped, but I think it will go back. It is the novelty of Discord that will likely wear off.

[grave ivy]

We are tracking this pretty closely, right now comments and topics are stable on the forums (no increase or decrease). My hope is that Discord will be helpful for people who currently don't post on the forums and that actually it might lead to more activity on both platforms, but we'll see how things go!

[edgy fulcrum]

You could check for yourself if you want in Meta Kaggle https://www.kaggle.com/datasets/kaggle/meta-kaggle

[fervent moat]

https://www.kaggle.com/datasets/yeoyunsianggeremie/s3e20-top-public-notebook-submissions. I created a dataset with most of the top public notebook submissions

however, from what I realised, in both S3E19 and S3E20, very few people are using it and most would rather add the actual notebook as a data source and use the submission.csv there to blend instead of using the prediction from this dataset

I am not sure why and is this dataset unnecessary work ? would appreciate some feedback

[fervent moat]

https://www.kaggle.com/competitions/playground-series-s3e20/discussion/431659 I have posted it here

[crude wind]

Agree with @west turret . I would prefer importing data from original sources not only because of the relative ease, but also because it acknowledges the work to the author of the output data

[fervent moat]

What topic are yall hoping for in the next playground ?

It probably won't be time series again lmao

[carmine kestrel]

Multi-class or anomalies

[fervent moat]

anomalies ? pls elaborate

[carmine kestrel]

https://towardsdatascience.com/how-to-use-machine-learning-for-anomaly-detection-and-condition-monitoring-6742f82900d7

[crude wind]

Sounds interesting!
I love anomalies detection 🥰

[fervent moat]

Wow massive shakeup

I chose the wrong submission

1st place went to 534th

I don’t actually have a submission in the top 50

[fervent moat]

Yup I definitely spent too much time trying to optimise non-ML approaches

😢

I believe many of us did too

[fervent moat]

Consequences of shakeup: we only see 1 official top 10 submission writeup