4 ms·
I was pretty surprised by how common this type of behavior is on Kaggle. I work in machine learning and data science, but I don't use Kaggle much because it qui
by bart_spoon 7y ago
I was pretty surprised by how common this type of behavior is on Kaggle. I work in machine learning and data science, but I don't use Kaggle much because it quickly became clear competitions boiled down to who could eek out the last hundredths of a percent in accuracy from models trained for weeks on multithousand dollar machines, and because the behavior described in the article was surprisingly common.
That said, the site is a fantastic resource for datasets. Lots of fantastic data uploaded both from old competitions and by the community.
- QuercusMax 7y agoIs a "multithousand dollar machine" supposed to be expensive? Any company with even half-decent resources should be able to put hundreds of thousands of dollars of hardware toward training a model.
- bart_spoon 7y agoIt is when the competitions are aimed at the machine learning practitioning public, and not companies themselves. If having a machine with a bare minimum of $3000 of GPU power is the entry point for having a competitive model, then most of these competitions aren't actually a matter of coming up with a unique and clever model. It's simply throwing the biggest neural network you can at it. Which is fine, but it definitely flies in the face of the perception of Kaggle.