NET Rankings | Syracusefan.com

NET Rankings

Gary Parrish‏ @GaryParrishCBS
I’m sure everything will be (mostly) fine in time. But this is a pretty embarrassing debut for the NCAA’s NET rankings. It looks like the computer is broken or something. It doesn’t reflect what it’s supposed to reflect.

If the options were to release this, or just hold it until there’s enough data to make it look sensible, the choice should’ve been to just hold it until there’s enough data to make it look sensible.
 
Andy Katz‏ @TheAndyKatz
Let the debate rage on. Very interesting numbers. But remember, this is one computer snapshot in time.
 
Media members need to sit this one out. They wanted a formula that gave mid majors a better chance, but now that one of their beloved blue bloods is ranked too low Gary Parrish is going to complain.
 
Whats so wrong with the top 10? The fact that it doesn't reflect the AP poll?
 
Without looking at the formula it must weigh in true road Ws heavily. OSU has two good ones of those already and aren’t allowing a lot of points as well.
 
Without looking at the formula it must weigh in true road Ws heavily. OSU has two good ones of those already and aren’t allowing a lot of points as well.
They actually haven't released the 'formula', just the inputs into the algorithm.

Interesting note - today's SU RPI is 138. SU's NET? 92.
1543274733162.gif
 
the way we started the season i dont know that mid 90s is wrong for us though
 
The main problem w this-

You either take last season into some account (which this doesn't) - or you wait until at least January to release the first one (which this doesn't)
 
Ohio State number 1 seems bizarre. So we’re playing the overall #1 this week - that alone should improve our NET, even if we lose, right?
 
Nate Silver needs to not weigh in on sports, let alone election results. Guy is too arrogant to admit when his modeling is off, which is quite often recently

Eh..i think he's on point here tho. And he was a sports guy first
 
Couple of observations:

-Info on this model is pretty bare besides the infographic in the tweet above, based on my quick googling
-There are fancy machine learning terms being thrown around, but it is an ensemble model, basically just adding weights to a group of other models
-It looks opaque now but someone will reverse engineer it soon, if it hasn't been done already
-It lacks priors, all teams start at zero, so it is going to produce unusual looking results until it gets enough data. In future years they are going to have to hold back until later
-RPI isn't dead, the "Team Value Index" which is the first and presumably largest input looks to just be a renamed RPI
-Net Efficiency, the second input, is a heavily trafficked NBA stat these days and the basis for KenPom. KenPom is heavily weighted for strength of opponent, this is not (or at least doesn't say it is). That is not a big deal in a closed system like the NBA where everyone plays everyone and there are common opponents, but with 350+ teams and major differences in quality it has issues.
-Scoring Margin, the last listed input, is something similar to Pythagorean Win Theorem, which works well in professional sports but has the same problem as Net Efficiency.
-Increasing the conference games is probably going to cause issues for this model going forward as teams exist more in their conference bubbles.
 
They should have released NET concurrently, but still only used RPI for at least a season or two so people could see how they differ and begin planning and scheduling accordingly. Seems insane that this is rolled out blind. Schools and conferences had no idea what the result was going to be.
 
-It lacks priors, all teams start at zero, so it is going to produce unusual looking results until it gets enough data. In future years they are going to have to hold back until later.

Exactly. Not sure why people are confused. It's November. The teams/rankings will look more normal in January.
 

Forum statistics

Threads
168,420
Messages
4,774,965
Members
5,949
Latest member
Laxmom2317

Online statistics

Members online
119
Guests online
883
Total visitors
1,002


Top Bottom