@goabiaryan Thanks for inviting me.

I agree with lost of what you say, esp that ML in production is not much about modeling, and much more about an end to end pipeline.

Kaggle zooms on a very specific and narrow part of such pipeline.

I'd still advocate Kaggle for one thing: it teaches you how to properly evaluate model performance.

Maybe things have changed, but most data scientists I've met who did not do kaggle use a flawed methodology for evaluating their model performance.

@goabiaryan way cooler than what?

I think I'll wait a bit to let dust settle. People seem to find me from other mastodon servers so far.

@PiotrGabrys

Thanks. 3 sucks, but do you find it was worth the effort?

@Riedl @djnavarro

Exactly what I was looking for: how to migrate to another server!