Talking Machines

Human conversations about machine learning on a beautifully designed ultra-minimalist website

Do you think it could raise statistical literacy and interest?

Some nice quotes:

K-nearest neighbor shouldn't outperform any other method unless you have duplicates in your data

-Claudia Perlich

If you find some data in your basement (as a marketing company) and analyse that it's not necessarily going to be actionable. It's better to ask: What are the actions you think you could be doing better if you had some piece of information, then we look for that information

-Claudia Perlich

Where else are you going to here that kind of stuff?!

Take head graduate students trying to decide what kind of research question to choose. Naive, descriptive analytics can be really dumb!

(Kaggle competitions lead to) overfitting to the noise

Take head prospective Kaggle competition hosters. You're engineering publication bias!