Even for science and medical applications, I am becoming weary of fine statistical modeling efforts, and believe that we should standardize on a handful of powerful and robust methods. An opinionated thread to give context for https://twitter.com/GaelVaroquaux/status/1223305762350534657 … 1/8
-
-
Ok a) (a nitpick) many ML models don't return calibrated expectations - I've seen this a lot with tree-based classifiers, but maybe that's software dependent. b) (not nitpick) what if you want to make a probability statement about your conditional expectation? ....
-
a) is very valid: calibrate expectation / probability is a weak point of machine learning, a genuine problem. b) needs more development in black-box model inspection, à la permutation importances. Both points call for research, but seem within reach.
- Još 6 drugih odgovora
Novi razgovor -
-
-
write a paper on this!
-
ML for estimation? Go on then write on this
Kraj razgovora
Novi razgovor -
-
-
*gentle push*
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
I'd be happy to see new methods in our pipeline, although I'd caution you not to overstate your case against small N data sets. For simple questions with a clear prediction and low variance, it's quite possible to get useful inference (i.e. it replicates) from a modest dataset
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Push!
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Would love a paper!
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
I think we could all benefit from your experience here. And it's very necessary to propose standards, especially for teams in industry. I try to standardize process in my team as much as possible, which is part of the fun in new areas like ML but we need lots of help from experts
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.