The corollary is that posts about highly-technical, cutting-edge techniques produce almost no discussion.
-
-
Prikaži ovu nitHvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
I enjoy the concern trolling comments below that say I am right but other things are also right. Peak stats twitter.
Prikaži ovu nitHvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Anonymous is not helping as they try to prove *asymptotic* property by finite number n=4 hand picked observations. Bayesian inference is almost always biased and uncalibrated and I agree that calibration is sometimes overrated, but I think that correctly used it's a useful tool.
-
Useful for evaluating decisions or just random number generators?
- Još 7 drugih odgovora
Novi razgovor -
-
-
Richard you seem to be confusing calibration-in-the-large with calibration-in-the-small. There's a huge literature on this especially in medical decision making.
-
What did I write that is wrong?
Kraj razgovora
Novi razgovor -
-
-
Sure, a calibrated (Probabilistically accurate) but imprecise prediction isn’t good, but neither is a precise but inaccurate one. A good prediction is both. I think Gneiting and others talk about good forecasts aiming to “maximize precision, subject to calibration”. 1/n
-
Imposing calibration above accuracy is similar to imposing unbiasedness, which I would agree is too strong a requirement, but relaxing that without being clear that one does that to help tradeoff with precision, or even only caring about precision doesn’t make sense to me. 2/2
- Još 1 odgovor
Novi razgovor -
-
-
Calibration is maybe underrated... been doing this for years and just recently started thinking re: calibration--and I have an excellent team...people just don't do it...and declare success with an "okay" ROC. But for things like clinical trial simulation, it's 100% necessary.
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
also, people are starting doing causal inference [sometimes incorrectly] using predictive models essentially as a non-parametric propensity score. This requires calibration otherwise you get spurious inference.
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.