APBRmetrics

eminence wrote: ↑Mon Jun 17, 2019 5:18 pm

RyanRiot wrote: ↑Mon Jun 17, 2019 4:41 pm Who does the model like this year?

The important questions

Likely to be better than consensus mock drafts: Brandon Clarke and Grant Williams.

Likely to be a little worse: R.J. Barrett and Darius Garland.

But ultimately the main takeaways for me are the importance of having an analytics process, and mining your unique treasure trove of historical scouting data that in reality teams tend to just ignore and throw away.

Maybe the importance of the questions lies in whether you can get this "historical" data and conclusions in a timely manner to those who can make use of it?
If the data is still coming out, your system has to be pretty responsive.

Here’s the model’s top few picks. Enjoy the draft tonight!

Player
Zion Williamson
De'Andre Hunter
Ja Morant
Jarrett Culver
Brandon Clarke
Jaxson Hayes
Coby White
R.J. Barrett
Bol Bol
Grant Williams
Cameron Reddish
Rui Hachimura
Cameron Johnson
Darius Garland

This is really neat. Only thing I would change is the dv. Win shares, EWA, and Wins Produced aren't the best measures of player performance. They're ok, but not the best.

This is a very cool idea. Do you have any retrodictions available so we can go "ooh, ahh" about the sleeper picks this system nailed?

kmedved wrote: ↑Tue Jun 25, 2019 7:36 pm This is a very cool idea. Do you have any retrodictions available so we can go "ooh, ahh" about the sleeper picks this system nailed?

Haha yes for sure. I’ll put them all up on a website soon and post the link here. I just found out I’ll be presenting a poster on it at NESSIS this year too, so if you’re there, come on by!

LOOCV is nice but it does not give us a sense for the variance in the model, and Random Forest can very easily be tuned to cheat and get "100% accuracy".

Do you have a version of this with test/train accuracy splits?