Thread by @radekosmulski

Two things the @kaggle @RSNA competition keeps reminding me of:

✅if you don't have a good validation set, you don't have anything
✅it doesn't matter what you think, really, if the experiments you run do not support your hypotheses

On the first point, I always keep pointing

people to this post by @math_rachel: fast.ai/2017/11/13/val…

It's hard to fathom how so much good information made its way into a single blog post. I claim sorcery.

If taking a more theoretical approach is your thing (it's okay, who am I to judge) the first couple of lectures

of edx.org/course/learnin… and the associated book smile.amazon.com/Learning-Data-… are great.