There is one topic that’s always my twitter Achilles heel: stats “vs” ML debates. Their existence frustrates me and I can never help myself. So, a few months back, I promised myself that the next time one came up, instead of entering the scrum, I’d just write a post. So.. 1/3https://twitter.com/jfutoma/status/1222837860803497984 …
-
-
To be clear, I’m under no illusion that this post will change anyone’s mind about anything. It was written quickly, and reasonable can disagree with many of my points. Rather, I hope it rids me of desire to waste time on toxic instantiations of this debate in the future. 3/3
Prikaži ovu nit -
PS Thanks
@JFutoma for prompting me to write it, and for commenting on my raw initial thoughts yesterday.Prikaži ovu nit
Kraj razgovora
Novi razgovor -
-
-
This controversy is funny from outside but nearby. As a computer scientist who knows nothing about stats, much less about machines learning*. I would have defined “machine learning” as “applied statistics for AI.” ————— *ML stands for Meta Language.
-
Yep, that’s not far off from where I define it. I even compare it to econometrics in this regard.
- Još 2 druga odgovora
Novi razgovor -
-
-
The only difference is that stat people view the data as a sample from a population while ML people don't. The odd thing, though, is that the MLers speak in sampling terms anyway, when talking about bias/variance and predicting new cases (presumably from the same distribution!).
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
LOL! Very nice, good job. Folks do struggle with cross-disciplinary stuff; silos are so safe ;-))
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
This is very well written! Thanks!
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Agreed on the missing historical context!
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
great writeup..that last part about using stat methods to run evals is golden. only thing i'd add is the fact that systems that ultimately hit prod are usually a combination like multi arm bandits. is that stat?ml? does it even matter?
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.
)