BearcatTalk.com

BearcatTalk.com (https://www.bearcattalk.com/index.php)
-   Bearcat Basketball (https://www.bearcattalk.com/forumdisplay.php?f=6)
-   -   new metric (https://www.bearcattalk.com/showthread.php?t=12108)

david aka the TYZ 08-22-2018 11:11 AM

new metric
 
https://sports.yahoo.com/ncaa-replaces-rpi-new-ncaa-tournament-selection-metric-154236270.html
NCAA replaces RPI with new March Madness selection metric


Just saw this. Very interesting and applies to about half our in season arguments. Curious to see this play out this year and in the future. Welcome to the term NET in all future rants on this board

cincrulz11 08-22-2018 12:23 PM

sounds worlds better than rpi. 10 point cap for win margin might be too low though.

sedziobs 08-22-2018 03:53 PM

This is a nice surprise. Looks like they're keeping the quadrant and team sheet system that was introduced last year. So we'll still have some basis for comparing resumes and projecting seeds.

david aka the TYZ 08-22-2018 04:05 PM

Yeah the 10 point cap seems low to me as well, but I do like the idea of it being capped. Not sure what a better cap would be. Maybe 20?

Wish they had more info on everything it factors in

david aka the TYZ 08-22-2018 04:14 PM

I will say if any team will be advantaged by a ten point cap it could be UC cause we don't seem to blow a lot of teams out we should. We'll beat a team by 13 that we coulda, shoulda, woulda beat by 30 more often than not it seems

GarradJ21 08-22-2018 05:38 PM

I wonder how this will affect the game itself. I mean a team has a ball up 8 and the other team decides not to foul since there’s only 5 seconds remaining. Does the team with the ball try to score?

sedziobs 11-27-2018 10:43 AM

Quote:

Originally Posted by cincrulz11 (Post 192667)
you'll probably have fun with NET. i guess we'll get to see if its any better than RPI was.


ive read a few things about it i dont like, capping wins at 10 points, and the quality of your opponent doesn't matter for the stats you put up (holding duke to .98 ppp is the same as holding arkansas-pine bluff to .98ppp) but im not sure if thats actually true or people misinterpreting the data.


if it is true, it is probably a big benefit to us considering our schedule. rpi sucked but was actually great for us last year.

I thought it would be good to have a separate thread for NET.

The RPI is dead. The quadrant system remains, but NET will be used instead of RPI. The main improvements are the inclusion of tempo and scoring margin. The main drawback is the 10 pt margin cap, which seems too low to me. I think 15 would be better.

The algorithm components are different, but efficiency is still adjusted for opponents in the final result. So .98ppp vs AK-PB is the same as .98ppp vs Duke in terms of efficiency, but it is adjusted separately to account for schedule strength (opponent and location). Kenpom does the same thing. His rankings are based on adjusted efficiency, which start with raw efficiency. RPI didn't account for efficiency at all. Overall I think it's a huge improvement.

The biggest challenge this season will be in predicting what will happen to the NET rankings as the season progresses. Last year, we were able to use a variety of RPI forecasting tools that ran monte carlo simulations based on the simple RPI formula. Now we are sort of flying blind, and will be limited to analyzing the rankings in the present.

Cincinnati is currently ranked #46. It's way too early to be very meaningful. Unlike Kenpom, NET does not include any pre-season projections. RPI did not either. In another month we'll have a better idea of how good teams, and especially conferences, are.

cincrulz11 11-27-2018 10:54 AM

Quote:

Originally Posted by sedziobs (Post 192678)
The algorithm components are different, but efficiency is still adjusted for opponents in the final result. So .98ppp vs AK-PB is the same as .98ppp vs Duke in terms of efficiency, but it is adjusted separately to account for schedule strength (opponent and location). Kenpom does the same thing. His rankings are based on adjusted efficiency, which start with raw efficiency. RPI didn't account for efficiency at all. Overall I think it's a huge improvement.


right but doesn't NET use 5 components to get your overall rating. so your game results factored in from strength of opponent is just 1 of the components. Then your raw efficiency, winning percentage, adjusted win percentage (adjusted for home, neutral, road), and scoring margin (capped at 10).



im not sure if each component is weighted the same or not. if they are all the same, that seems to benefit teams that beat up bad opponents.

sedziobs 11-27-2018 11:01 AM

Quote:

Originally Posted by cincrulz11 (Post 192680)
right but doesn't NET use 5 components to get your overall rating. so your game results factored in from strength of opponent is just 1 of the components. Then your raw efficiency, winning percentage, adjusted win percentage (adjusted for home, neutral, road), and scoring margin (capped at 10).

im not sure if each component is weighted the same or not. if they are all the same, that seems to benefit teams that beat up bad opponents.

We don't know what the weights are, but they aren't all the same. They used a machine learning algorithm to develop a model that determined what weights gave the best hindcast results. Assuming it's properly calibrated, this would be an improvement over even Kenpom. I'm a little worried that it might be overfitted (meaning it predicts past results with past data really well, but not necessarily future ones - it could be overconfident that it can predict randomness). It's possible it will adjust the weightings in real time to correct itself.

cincrulz11 11-27-2018 11:04 AM

Quote:

Originally Posted by sedziobs (Post 192681)
We don't know what the weights are, but they aren't all the same. They used a machine learning algorithm to develop a model that determined what weights gave the best hindcast results. Assuming it's properly calibrated, this would be an improvement over even Kenpom. I'm a little worried that it might be overfitted (meaning it predicts past results with past data really well, but not necessarily future ones - it could be overconfident that it can predict randomness). It's possible it will adjust the weightings in real time to correct itself.


well then that sounds much better than the ranting nate silver has been doing on twitter.


All times are GMT -5. The time now is 03:29 PM.

Powered by vBulletin® Copyright ©2000 - 2024, Jelsoft Enterprises Ltd.