Read on Twitter
Any of you all want to share something you learned while analyzing COVID / Coronavirus data? Were you surprised? Did something defy your intuition? I caught some shit for encouraging us all to think in detail about this. Have you thought? Learn anything? https://twitter.com/CMastication/status/1238075732573683713?s=20
I'll share a few things I've learned:
* It's easy to get distracted by very specific things like, "predict the mortality number on day X" That rarely matters. Usually we're trying to answer some other question but we get suckered into "predict Y"... but we should focus on the Q:
* It's easy to get distracted by very specific things like, "predict the mortality number on day X" That rarely matters. Usually we're trying to answer some other question but we get suckered into "predict Y"... but we should focus on the Q:
If the question is "should we covert our 50 bed hospital wing into an ICU?" then the answer does not depend on predicting the mort number on a given day. It's more like "what do we have to believe to think our ICU will be overrun?" ... that's a MUCH easier question.
This is (yet again) my issue with data science competitions. In order to judge they have to reduce things to some numerical "goodness of fit" metric. And the top 100 are separated by 0.000035% (or something). Meanwhile what matters in practice is reducing a question to an action
and what we need in practice is to make a much more coarse choice: "will X happen".. or "how many times can we do Y?" ... these often need speed, not infinite precision. The *killer* skill is knowing *what decision are we actually driving*....
I'm clearly thinking about "using analysis to drive discrete business decisions" which is *TOTALLY DIFFERENT* from "building a machine learning tool to assist in XYZ" .. My comments are biased by my experiences in corporate decision making. Not from "AI consumer products"